blob: 044170165402f0c2806a6038289b5f545a485673 [file] [log] [blame]
Jens Axboe2b188cc2019-01-07 10:46:33 -07001// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared application/kernel submission and completion ring pairs, for
4 * supporting fast/efficient IO.
5 *
6 * A note on the read/write ordering memory barriers that are matched between
Stefan Bühler1e84b972019-04-24 23:54:16 +02007 * the application and kernel side.
8 *
9 * After the application reads the CQ ring tail, it must use an
10 * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses
11 * before writing the tail (using smp_load_acquire to read the tail will
12 * do). It also needs a smp_mb() before updating CQ head (ordering the
13 * entry load(s) with the head store), pairing with an implicit barrier
14 * through a control-dependency in io_get_cqring (smp_store_release to
15 * store head will do). Failure to do so could lead to reading invalid
16 * CQ entries.
17 *
18 * Likewise, the application must use an appropriate smp_wmb() before
19 * writing the SQ tail (ordering SQ entry stores with the tail store),
20 * which pairs with smp_load_acquire in io_get_sqring (smp_store_release
21 * to store the tail will do). And it needs a barrier ordering the SQ
22 * head load before writing new SQ entries (smp_load_acquire to read
23 * head will do).
24 *
25 * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application
26 * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after*
27 * updating the SQ tail; a full memory barrier smp_mb() is needed
28 * between.
Jens Axboe2b188cc2019-01-07 10:46:33 -070029 *
30 * Also see the examples in the liburing library:
31 *
32 * git://git.kernel.dk/liburing
33 *
34 * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens
35 * from data shared between the kernel and application. This is done both
36 * for ordering purposes, but also to ensure that once a value is loaded from
37 * data that the application could potentially modify, it remains stable.
38 *
39 * Copyright (C) 2018-2019 Jens Axboe
Christoph Hellwigc992fe22019-01-11 09:43:02 -070040 * Copyright (c) 2018-2019 Christoph Hellwig
Jens Axboe2b188cc2019-01-07 10:46:33 -070041 */
42#include <linux/kernel.h>
43#include <linux/init.h>
44#include <linux/errno.h>
45#include <linux/syscalls.h>
46#include <linux/compat.h>
Jens Axboe52de1fe2020-02-27 10:15:42 -070047#include <net/compat.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070048#include <linux/refcount.h>
49#include <linux/uio.h>
Pavel Begunkov6b47ee62020-01-18 20:22:41 +030050#include <linux/bits.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070051
52#include <linux/sched/signal.h>
53#include <linux/fs.h>
54#include <linux/file.h>
55#include <linux/fdtable.h>
56#include <linux/mm.h>
57#include <linux/mman.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070058#include <linux/percpu.h>
59#include <linux/slab.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070060#include <linux/blkdev.h>
Jens Axboeedafcce2019-01-09 09:16:05 -070061#include <linux/bvec.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070062#include <linux/net.h>
63#include <net/sock.h>
64#include <net/af_unix.h>
Jens Axboe6b063142019-01-10 22:13:58 -070065#include <net/scm.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070066#include <linux/anon_inodes.h>
67#include <linux/sched/mm.h>
68#include <linux/uaccess.h>
69#include <linux/nospec.h>
Jens Axboeedafcce2019-01-09 09:16:05 -070070#include <linux/sizes.h>
71#include <linux/hugetlb.h>
Jens Axboeaa4c3962019-11-29 10:14:00 -070072#include <linux/highmem.h>
Jens Axboe15b71ab2019-12-11 11:20:36 -070073#include <linux/namei.h>
74#include <linux/fsnotify.h>
Jens Axboe4840e412019-12-25 22:03:45 -070075#include <linux/fadvise.h>
Jens Axboe3e4827b2020-01-08 15:18:09 -070076#include <linux/eventpoll.h>
Pavel Begunkov7d67af22020-02-24 11:32:45 +030077#include <linux/splice.h>
Jens Axboeb41e9852020-02-17 09:52:41 -070078#include <linux/task_work.h>
Jens Axboebcf5a062020-05-22 09:24:42 -060079#include <linux/pagemap.h>
Jens Axboe0f212202020-09-13 13:09:39 -060080#include <linux/io_uring.h>
Jens Axboee4b4a132021-03-01 18:36:25 -070081#include <linux/freezer.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070082
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020083#define CREATE_TRACE_POINTS
84#include <trace/events/io_uring.h>
85
Jens Axboe2b188cc2019-01-07 10:46:33 -070086#include <uapi/linux/io_uring.h>
87
88#include "internal.h"
Jens Axboe561fb042019-10-24 07:25:42 -060089#include "io-wq.h"
Jens Axboe2b188cc2019-01-07 10:46:33 -070090
Daniel Xu5277dea2019-09-14 14:23:45 -070091#define IORING_MAX_ENTRIES 32768
Jens Axboe33a107f2019-10-04 12:10:03 -060092#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
Jens Axboe65e19f52019-10-26 07:20:21 -060093
94/*
95 * Shift of 9 is 512 entries, or exactly one page on 64-bit archs
96 */
97#define IORING_FILE_TABLE_SHIFT 9
98#define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT)
99#define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1)
100#define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE)
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200101#define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \
102 IORING_REGISTER_LAST + IORING_OP_LAST)
Jens Axboe2b188cc2019-01-07 10:46:33 -0700103
Pavel Begunkovb16fed662021-02-18 18:29:40 +0000104#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
105 IOSQE_IO_HARDLINK | IOSQE_ASYNC | \
106 IOSQE_BUFFER_SELECT)
107
Jens Axboe2b188cc2019-01-07 10:46:33 -0700108struct io_uring {
109 u32 head ____cacheline_aligned_in_smp;
110 u32 tail ____cacheline_aligned_in_smp;
111};
112
Stefan Bühler1e84b972019-04-24 23:54:16 +0200113/*
Hristo Venev75b28af2019-08-26 17:23:46 +0000114 * This data is shared with the application through the mmap at offsets
115 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200116 *
117 * The offsets to the member fields are published through struct
118 * io_sqring_offsets when calling io_uring_setup.
119 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000120struct io_rings {
Stefan Bühler1e84b972019-04-24 23:54:16 +0200121 /*
122 * Head and tail offsets into the ring; the offsets need to be
123 * masked to get valid indices.
124 *
Hristo Venev75b28af2019-08-26 17:23:46 +0000125 * The kernel controls head of the sq ring and the tail of the cq ring,
126 * and the application controls tail of the sq ring and the head of the
127 * cq ring.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200128 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000129 struct io_uring sq, cq;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200130 /*
Hristo Venev75b28af2019-08-26 17:23:46 +0000131 * Bitmasks to apply to head and tail offsets (constant, equals
Stefan Bühler1e84b972019-04-24 23:54:16 +0200132 * ring_entries - 1)
133 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000134 u32 sq_ring_mask, cq_ring_mask;
135 /* Ring sizes (constant, power of 2) */
136 u32 sq_ring_entries, cq_ring_entries;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200137 /*
138 * Number of invalid entries dropped by the kernel due to
139 * invalid index stored in array
140 *
141 * Written by the kernel, shouldn't be modified by the
142 * application (i.e. get number of "new events" by comparing to
143 * cached value).
144 *
145 * After a new SQ head value was read by the application this
146 * counter includes all submissions that were dropped reaching
147 * the new SQ head (and possibly more).
148 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000149 u32 sq_dropped;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200150 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200151 * Runtime SQ flags
Stefan Bühler1e84b972019-04-24 23:54:16 +0200152 *
153 * Written by the kernel, shouldn't be modified by the
154 * application.
155 *
156 * The application needs a full memory barrier before checking
157 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
158 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000159 u32 sq_flags;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200160 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200161 * Runtime CQ flags
162 *
163 * Written by the application, shouldn't be modified by the
164 * kernel.
165 */
166 u32 cq_flags;
167 /*
Stefan Bühler1e84b972019-04-24 23:54:16 +0200168 * Number of completion events lost because the queue was full;
169 * this should be avoided by the application by making sure
LimingWu0b4295b2019-12-05 20:18:18 +0800170 * there are not more requests pending than there is space in
Stefan Bühler1e84b972019-04-24 23:54:16 +0200171 * the completion queue.
172 *
173 * Written by the kernel, shouldn't be modified by the
174 * application (i.e. get number of "new events" by comparing to
175 * cached value).
176 *
177 * As completion events come in out of order this counter is not
178 * ordered with any other data.
179 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000180 u32 cq_overflow;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200181 /*
182 * Ring buffer of completion events.
183 *
184 * The kernel writes completion events fresh every time they are
185 * produced, so the application is allowed to modify pending
186 * entries.
187 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000188 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700189};
190
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000191enum io_uring_cmd_flags {
192 IO_URING_F_NONBLOCK = 1,
Pavel Begunkov889fca72021-02-10 00:03:09 +0000193 IO_URING_F_COMPLETE_DEFER = 2,
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000194};
195
Jens Axboeedafcce2019-01-09 09:16:05 -0700196struct io_mapped_ubuf {
197 u64 ubuf;
198 size_t len;
199 struct bio_vec *bvec;
200 unsigned int nr_bvecs;
Jens Axboede293932020-09-17 16:19:16 -0600201 unsigned long acct_pages;
Jens Axboeedafcce2019-01-09 09:16:05 -0700202};
203
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000204struct io_ring_ctx;
205
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000206struct io_rsrc_put {
207 struct list_head list;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000208 union {
209 void *rsrc;
210 struct file *file;
211 };
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000212};
213
214struct fixed_rsrc_table {
Jens Axboe65e19f52019-10-26 07:20:21 -0600215 struct file **files;
Jens Axboe31b51512019-01-18 22:56:34 -0700216};
217
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000218struct fixed_rsrc_ref_node {
Xiaoguang Wang05589552020-03-31 14:05:18 +0800219 struct percpu_ref refs;
220 struct list_head node;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000221 struct list_head rsrc_list;
222 struct fixed_rsrc_data *rsrc_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000223 void (*rsrc_put)(struct io_ring_ctx *ctx,
224 struct io_rsrc_put *prsrc);
Jens Axboe4a38aed22020-05-14 17:21:15 -0600225 struct llist_node llist;
Pavel Begunkove2978222020-11-18 14:56:26 +0000226 bool done;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800227};
228
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000229struct fixed_rsrc_data {
230 struct fixed_rsrc_table *table;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700231 struct io_ring_ctx *ctx;
232
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000233 struct fixed_rsrc_ref_node *node;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700234 struct percpu_ref refs;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700235 struct completion done;
Hao Xu8bad28d2021-02-19 17:19:36 +0800236 bool quiesce;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700237};
238
Jens Axboe5a2e7452020-02-23 16:23:11 -0700239struct io_buffer {
240 struct list_head list;
241 __u64 addr;
242 __s32 len;
243 __u16 bid;
244};
245
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200246struct io_restriction {
247 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
248 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
249 u8 sqe_flags_allowed;
250 u8 sqe_flags_required;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +0200251 bool registered;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200252};
253
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700254enum {
255 IO_SQ_THREAD_SHOULD_STOP = 0,
256 IO_SQ_THREAD_SHOULD_PARK,
257};
258
Jens Axboe534ca6d2020-09-02 13:52:19 -0600259struct io_sq_data {
260 refcount_t refs;
Jens Axboe69fb2132020-09-14 11:16:23 -0600261 struct mutex lock;
262
263 /* ctx's that are using this sqd */
264 struct list_head ctx_list;
265 struct list_head ctx_new_list;
266 struct mutex ctx_lock;
267
Jens Axboe534ca6d2020-09-02 13:52:19 -0600268 struct task_struct *thread;
269 struct wait_queue_head wait;
Xiaoguang Wang08369242020-11-03 14:15:59 +0800270
271 unsigned sq_thread_idle;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700272 int sq_cpu;
273 pid_t task_pid;
274
275 unsigned long state;
276 struct completion startup;
277 struct completion completion;
278 struct completion exited;
Jens Axboe534ca6d2020-09-02 13:52:19 -0600279};
280
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000281#define IO_IOPOLL_BATCH 8
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000282#define IO_COMPL_BATCH 32
Pavel Begunkov6ff119a2021-02-10 00:03:18 +0000283#define IO_REQ_CACHE_SIZE 32
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000284#define IO_REQ_ALLOC_BATCH 8
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000285
286struct io_comp_state {
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000287 struct io_kiocb *reqs[IO_COMPL_BATCH];
Jens Axboe1b4c3512021-02-10 00:03:19 +0000288 unsigned int nr;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700289 unsigned int locked_free_nr;
290 /* inline/task_work completion list, under ->uring_lock */
Jens Axboe1b4c3512021-02-10 00:03:19 +0000291 struct list_head free_list;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700292 /* IRQ completion list, under ->completion_lock */
293 struct list_head locked_free_list;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000294};
295
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000296struct io_submit_link {
297 struct io_kiocb *head;
298 struct io_kiocb *last;
299};
300
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000301struct io_submit_state {
302 struct blk_plug plug;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000303 struct io_submit_link link;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000304
305 /*
306 * io_kiocb alloc cache
307 */
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000308 void *reqs[IO_REQ_CACHE_SIZE];
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000309 unsigned int free_reqs;
310
311 bool plug_started;
312
313 /*
314 * Batch completion logic
315 */
316 struct io_comp_state comp;
317
318 /*
319 * File reference cache
320 */
321 struct file *file;
322 unsigned int fd;
323 unsigned int file_refs;
324 unsigned int ios_left;
325};
326
Jens Axboe2b188cc2019-01-07 10:46:33 -0700327struct io_ring_ctx {
328 struct {
329 struct percpu_ref refs;
330 } ____cacheline_aligned_in_smp;
331
332 struct {
333 unsigned int flags;
Randy Dunlape1d85332020-02-05 20:57:10 -0800334 unsigned int compat: 1;
Randy Dunlape1d85332020-02-05 20:57:10 -0800335 unsigned int cq_overflow_flushed: 1;
336 unsigned int drain_next: 1;
337 unsigned int eventfd_async: 1;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200338 unsigned int restricted: 1;
Jens Axboe5f3f26f2021-02-25 10:17:46 -0700339 unsigned int sqo_exec: 1;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700340
Hristo Venev75b28af2019-08-26 17:23:46 +0000341 /*
342 * Ring buffer of indices into array of io_uring_sqe, which is
343 * mmapped by the application using the IORING_OFF_SQES offset.
344 *
345 * This indirection could e.g. be used to assign fixed
346 * io_uring_sqe entries to operations and only submit them to
347 * the queue when needed.
348 *
349 * The kernel modifies neither the indices array nor the entries
350 * array.
351 */
352 u32 *sq_array;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700353 unsigned cached_sq_head;
354 unsigned sq_entries;
355 unsigned sq_mask;
Jens Axboe6c271ce2019-01-10 11:22:30 -0700356 unsigned sq_thread_idle;
Jens Axboe498ccd92019-10-25 10:04:25 -0600357 unsigned cached_sq_dropped;
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +0100358 unsigned cached_cq_overflow;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700359 unsigned long sq_check_overflow;
Jens Axboede0617e2019-04-06 21:51:27 -0600360
Jens Axboee9418942021-02-19 12:33:30 -0700361 /* hashed buffered write serialization */
362 struct io_wq_hash *hash_map;
363
Jens Axboede0617e2019-04-06 21:51:27 -0600364 struct list_head defer_list;
Jens Axboe5262f562019-09-17 12:26:57 -0600365 struct list_head timeout_list;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -0700366 struct list_head cq_overflow_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700367
Jens Axboead3eb2c2019-12-18 17:12:20 -0700368 struct io_uring_sqe *sq_sqes;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700369 } ____cacheline_aligned_in_smp;
370
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700371 struct {
372 struct mutex uring_lock;
373 wait_queue_head_t wait;
374 } ____cacheline_aligned_in_smp;
375
376 struct io_submit_state submit_state;
377
Hristo Venev75b28af2019-08-26 17:23:46 +0000378 struct io_rings *rings;
379
Jens Axboe2aede0e2020-09-14 10:45:53 -0600380 /* Only used for accounting purposes */
381 struct mm_struct *mm_account;
382
Jens Axboe534ca6d2020-09-02 13:52:19 -0600383 struct io_sq_data *sq_data; /* if using sq thread polling */
384
Jens Axboe90554202020-09-03 12:12:41 -0600385 struct wait_queue_head sqo_sq_wait;
Jens Axboe69fb2132020-09-14 11:16:23 -0600386 struct list_head sqd_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700387
Jens Axboe6b063142019-01-10 22:13:58 -0700388 /*
389 * If used, fixed file set. Writers must ensure that ->refs is dead,
390 * readers must ensure that ->refs is alive as long as the file* is
391 * used. Only updated through io_uring_register(2).
392 */
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000393 struct fixed_rsrc_data *file_data;
Jens Axboe6b063142019-01-10 22:13:58 -0700394 unsigned nr_user_files;
395
Jens Axboeedafcce2019-01-09 09:16:05 -0700396 /* if used, fixed mapped user buffers */
397 unsigned nr_user_bufs;
398 struct io_mapped_ubuf *user_bufs;
399
Jens Axboe2b188cc2019-01-07 10:46:33 -0700400 struct user_struct *user;
401
Jens Axboe0f158b42020-05-14 17:18:39 -0600402 struct completion ref_comp;
403 struct completion sq_thread_comp;
Jens Axboe206aefd2019-11-07 18:27:42 -0700404
405#if defined(CONFIG_UNIX)
406 struct socket *ring_sock;
407#endif
408
Jens Axboe5a2e7452020-02-23 16:23:11 -0700409 struct idr io_buffer_idr;
410
Jens Axboe071698e2020-01-28 10:04:42 -0700411 struct idr personality_idr;
412
Jens Axboe206aefd2019-11-07 18:27:42 -0700413 struct {
414 unsigned cached_cq_tail;
415 unsigned cq_entries;
416 unsigned cq_mask;
417 atomic_t cq_timeouts;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -0500418 unsigned cq_last_tm_flush;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700419 unsigned long cq_check_overflow;
Jens Axboe206aefd2019-11-07 18:27:42 -0700420 struct wait_queue_head cq_wait;
421 struct fasync_struct *cq_fasync;
422 struct eventfd_ctx *cq_ev_fd;
423 } ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700424
425 struct {
Jens Axboe2b188cc2019-01-07 10:46:33 -0700426 spinlock_t completion_lock;
Jens Axboee94f1412019-12-19 12:06:02 -0700427
Jens Axboedef596e2019-01-09 08:59:42 -0700428 /*
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300429 * ->iopoll_list is protected by the ctx->uring_lock for
Jens Axboedef596e2019-01-09 08:59:42 -0700430 * io_uring instances that don't use IORING_SETUP_SQPOLL.
431 * For SQPOLL, only the single threaded io_sq_thread() will
432 * manipulate the list, hence no extra locking is needed there.
433 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300434 struct list_head iopoll_list;
Jens Axboe78076bb2019-12-04 19:56:40 -0700435 struct hlist_head *cancel_hash;
436 unsigned cancel_hash_bits;
Jens Axboee94f1412019-12-19 12:06:02 -0700437 bool poll_multi_file;
Jens Axboefcb323c2019-10-24 12:39:47 -0600438
439 spinlock_t inflight_lock;
440 struct list_head inflight_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700441 } ____cacheline_aligned_in_smp;
Jens Axboe85faa7b2020-04-09 18:14:00 -0600442
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000443 struct delayed_work rsrc_put_work;
444 struct llist_head rsrc_put_llist;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +0000445 struct list_head rsrc_ref_list;
446 spinlock_t rsrc_ref_lock;
Jens Axboe4a38aed22020-05-14 17:21:15 -0600447
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200448 struct io_restriction restrictions;
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700449
Jens Axboe7c25c0d2021-02-16 07:17:00 -0700450 /* exit task_work */
451 struct callback_head *exit_task_work;
452
Jens Axboee9418942021-02-19 12:33:30 -0700453 struct wait_queue_head hash_wait;
454
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700455 /* Keep this last, we don't need it for the fast path */
456 struct work_struct exit_work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700457};
458
Jens Axboe09bb8392019-03-13 12:39:28 -0600459/*
460 * First field must be the file pointer in all the
461 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
462 */
Jens Axboe221c5eb2019-01-17 09:41:58 -0700463struct io_poll_iocb {
464 struct file *file;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000465 struct wait_queue_head *head;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700466 __poll_t events;
Jens Axboe8c838782019-03-12 15:48:16 -0600467 bool done;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700468 bool canceled;
Jens Axboe392edb42019-12-09 17:52:20 -0700469 struct wait_queue_entry wait;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700470};
471
Pavel Begunkov018043b2020-10-27 23:17:18 +0000472struct io_poll_remove {
473 struct file *file;
474 u64 addr;
475};
476
Jens Axboeb5dba592019-12-11 14:02:38 -0700477struct io_close {
478 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -0700479 int fd;
480};
481
Jens Axboead8a48a2019-11-15 08:49:11 -0700482struct io_timeout_data {
483 struct io_kiocb *req;
484 struct hrtimer timer;
485 struct timespec64 ts;
486 enum hrtimer_mode mode;
487};
488
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700489struct io_accept {
490 struct file *file;
491 struct sockaddr __user *addr;
492 int __user *addr_len;
493 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600494 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700495};
496
497struct io_sync {
498 struct file *file;
499 loff_t len;
500 loff_t off;
501 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700502 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700503};
504
Jens Axboefbf23842019-12-17 18:45:56 -0700505struct io_cancel {
506 struct file *file;
507 u64 addr;
508};
509
Jens Axboeb29472e2019-12-17 18:50:29 -0700510struct io_timeout {
511 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300512 u32 off;
513 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300514 struct list_head list;
Pavel Begunkov90cd7e42020-10-27 23:25:36 +0000515 /* head of the link, used by linked timeouts only */
516 struct io_kiocb *head;
Jens Axboeb29472e2019-12-17 18:50:29 -0700517};
518
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100519struct io_timeout_rem {
520 struct file *file;
521 u64 addr;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000522
523 /* timeout update */
524 struct timespec64 ts;
525 u32 flags;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100526};
527
Jens Axboe9adbd452019-12-20 08:45:55 -0700528struct io_rw {
529 /* NOTE: kiocb has the file as the first member, so don't do it here */
530 struct kiocb kiocb;
531 u64 addr;
532 u64 len;
533};
534
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700535struct io_connect {
536 struct file *file;
537 struct sockaddr __user *addr;
538 int addr_len;
539};
540
Jens Axboee47293f2019-12-20 08:58:21 -0700541struct io_sr_msg {
542 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700543 union {
Pavel Begunkov270a5942020-07-12 20:41:04 +0300544 struct user_msghdr __user *umsg;
Jens Axboefddafac2020-01-04 20:19:44 -0700545 void __user *buf;
546 };
Jens Axboee47293f2019-12-20 08:58:21 -0700547 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700548 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700549 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700550 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700551};
552
Jens Axboe15b71ab2019-12-11 11:20:36 -0700553struct io_open {
554 struct file *file;
555 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700556 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700557 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600558 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700559};
560
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000561struct io_rsrc_update {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700562 struct file *file;
563 u64 arg;
564 u32 nr_args;
565 u32 offset;
566};
567
Jens Axboe4840e412019-12-25 22:03:45 -0700568struct io_fadvise {
569 struct file *file;
570 u64 offset;
571 u32 len;
572 u32 advice;
573};
574
Jens Axboec1ca7572019-12-25 22:18:28 -0700575struct io_madvise {
576 struct file *file;
577 u64 addr;
578 u32 len;
579 u32 advice;
580};
581
Jens Axboe3e4827b2020-01-08 15:18:09 -0700582struct io_epoll {
583 struct file *file;
584 int epfd;
585 int op;
586 int fd;
587 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700588};
589
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300590struct io_splice {
591 struct file *file_out;
592 struct file *file_in;
593 loff_t off_out;
594 loff_t off_in;
595 u64 len;
596 unsigned int flags;
597};
598
Jens Axboeddf0322d2020-02-23 16:41:33 -0700599struct io_provide_buf {
600 struct file *file;
601 __u64 addr;
602 __s32 len;
603 __u32 bgid;
604 __u16 nbufs;
605 __u16 bid;
606};
607
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700608struct io_statx {
609 struct file *file;
610 int dfd;
611 unsigned int mask;
612 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700613 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700614 struct statx __user *buffer;
615};
616
Jens Axboe36f4fa62020-09-05 11:14:22 -0600617struct io_shutdown {
618 struct file *file;
619 int how;
620};
621
Jens Axboe80a261f2020-09-28 14:23:58 -0600622struct io_rename {
623 struct file *file;
624 int old_dfd;
625 int new_dfd;
626 struct filename *oldpath;
627 struct filename *newpath;
628 int flags;
629};
630
Jens Axboe14a11432020-09-28 14:27:37 -0600631struct io_unlink {
632 struct file *file;
633 int dfd;
634 int flags;
635 struct filename *filename;
636};
637
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300638struct io_completion {
639 struct file *file;
640 struct list_head list;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +0300641 int cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300642};
643
Jens Axboef499a022019-12-02 16:28:46 -0700644struct io_async_connect {
645 struct sockaddr_storage address;
646};
647
Jens Axboe03b12302019-12-02 18:50:25 -0700648struct io_async_msghdr {
649 struct iovec fast_iov[UIO_FASTIOV];
Pavel Begunkov257e84a2021-02-05 00:58:00 +0000650 /* points to an allocated iov, if NULL we use fast_iov instead */
651 struct iovec *free_iov;
Jens Axboe03b12302019-12-02 18:50:25 -0700652 struct sockaddr __user *uaddr;
653 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700654 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700655};
656
Jens Axboef67676d2019-12-02 11:03:47 -0700657struct io_async_rw {
658 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600659 const struct iovec *free_iovec;
660 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600661 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600662 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700663};
664
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300665enum {
666 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
667 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
668 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
669 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
670 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700671 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300672
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300673 REQ_F_FAIL_LINK_BIT,
674 REQ_F_INFLIGHT_BIT,
675 REQ_F_CUR_POS_BIT,
676 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300677 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300678 REQ_F_ISREG_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300679 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700680 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700681 REQ_F_BUFFER_SELECTED_BIT,
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600682 REQ_F_NO_FILE_TABLE_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100683 REQ_F_LTIMEOUT_ACTIVE_BIT,
Pavel Begunkove342c802021-01-19 13:32:47 +0000684 REQ_F_COMPLETE_INLINE_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700685
686 /* not a real bit, just to check we're not overflowing the space */
687 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300688};
689
690enum {
691 /* ctx owns file */
692 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
693 /* drain existing IO first */
694 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
695 /* linked sqes */
696 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
697 /* doesn't sever on completion < 0 */
698 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
699 /* IOSQE_ASYNC */
700 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700701 /* IOSQE_BUFFER_SELECT */
702 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300703
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300704 /* fail rest of links */
705 REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT),
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +0000706 /* on inflight list, should be cancelled and waited on exit reliably */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300707 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
708 /* read/write uses file position */
709 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
710 /* must not punt to workers */
711 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100712 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300713 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300714 /* regular file */
715 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300716 /* needs cleanup */
717 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700718 /* already went through poll handler */
719 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700720 /* buffer already selected */
721 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600722 /* doesn't need file table for this request */
723 REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100724 /* linked timeout is active, i.e. prepared by link's head */
725 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Pavel Begunkove342c802021-01-19 13:32:47 +0000726 /* completion is deferred through io_comp_state */
727 REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700728};
729
730struct async_poll {
731 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600732 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300733};
734
Jens Axboe7cbf1722021-02-10 00:03:20 +0000735struct io_task_work {
736 struct io_wq_work_node node;
737 task_work_func_t func;
738};
739
Jens Axboe09bb8392019-03-13 12:39:28 -0600740/*
741 * NOTE! Each of the iocb union members has the file pointer
742 * as the first entry in their struct definition. So you can
743 * access the file pointer through any of the sub-structs,
744 * or directly as just 'ki_filp' in this struct.
745 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700746struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700747 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600748 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700749 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700750 struct io_poll_iocb poll;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000751 struct io_poll_remove poll_remove;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700752 struct io_accept accept;
753 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700754 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700755 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100756 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700757 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700758 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700759 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700760 struct io_close close;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000761 struct io_rsrc_update rsrc_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700762 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700763 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700764 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300765 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700766 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700767 struct io_statx statx;
Jens Axboe36f4fa62020-09-05 11:14:22 -0600768 struct io_shutdown shutdown;
Jens Axboe80a261f2020-09-28 14:23:58 -0600769 struct io_rename rename;
Jens Axboe14a11432020-09-28 14:27:37 -0600770 struct io_unlink unlink;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300771 /* use only after cleaning per-op data, see io_clean_op() */
772 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700773 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700774
Jens Axboee8c2bc12020-08-15 18:44:09 -0700775 /* opcode allocated if it needs to store data for async defer */
776 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700777 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800778 /* polled IO has completed */
779 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700780
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700781 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300782 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700783
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300784 struct io_ring_ctx *ctx;
785 unsigned int flags;
786 refcount_t refs;
787 struct task_struct *task;
788 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700789
Pavel Begunkovf2f87372020-10-27 23:25:37 +0000790 struct io_kiocb *link;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000791 struct percpu_ref *fixed_rsrc_refs;
Jens Axboed7718a92020-02-14 22:23:12 -0700792
Pavel Begunkovd21ffe72020-07-13 23:37:10 +0300793 /*
794 * 1. used with ctx->iopoll_list with reads/writes
795 * 2. to track reqs with ->files (see io_op_def::file_table)
796 */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300797 struct list_head inflight_entry;
Jens Axboe7cbf1722021-02-10 00:03:20 +0000798 union {
799 struct io_task_work io_task_work;
800 struct callback_head task_work;
801 };
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300802 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
803 struct hlist_node hash_node;
804 struct async_poll *apoll;
805 struct io_wq_work work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700806};
807
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300808struct io_defer_entry {
809 struct list_head list;
810 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300811 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300812};
813
Jens Axboed3656342019-12-18 09:50:26 -0700814struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700815 /* needs req->file assigned */
816 unsigned needs_file : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700817 /* hash wq insertion if file is a regular file */
818 unsigned hash_reg_file : 1;
819 /* unbound wq insertion if file is a non-regular file */
820 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700821 /* opcode is not supported by this kernel */
822 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700823 /* set if opcode supports polled "wait" */
824 unsigned pollin : 1;
825 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700826 /* op supports buffer selection */
827 unsigned buffer_select : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700828 /* must always have async data allocated */
829 unsigned needs_async_data : 1;
Jens Axboe27926b62020-10-28 09:33:23 -0600830 /* should block plug */
831 unsigned plug : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700832 /* size of async data needed, if any */
833 unsigned short async_size;
Jens Axboed3656342019-12-18 09:50:26 -0700834};
835
Jens Axboe09186822020-10-13 15:01:40 -0600836static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300837 [IORING_OP_NOP] = {},
838 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700839 .needs_file = 1,
840 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700841 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700842 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700843 .needs_async_data = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600844 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700845 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700846 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300847 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700848 .needs_file = 1,
849 .hash_reg_file = 1,
850 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700851 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700852 .needs_async_data = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600853 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700854 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700855 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300856 [IORING_OP_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700857 .needs_file = 1,
858 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300859 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700860 .needs_file = 1,
861 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700862 .pollin = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600863 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700864 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700865 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300866 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700867 .needs_file = 1,
868 .hash_reg_file = 1,
869 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700870 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600871 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700872 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700873 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300874 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700875 .needs_file = 1,
876 .unbound_nonreg_file = 1,
877 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300878 [IORING_OP_POLL_REMOVE] = {},
879 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700880 .needs_file = 1,
881 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300882 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700883 .needs_file = 1,
884 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700885 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700886 .needs_async_data = 1,
887 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700888 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300889 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700890 .needs_file = 1,
891 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700892 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700893 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700894 .needs_async_data = 1,
895 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700896 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300897 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700898 .needs_async_data = 1,
899 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700900 },
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000901 [IORING_OP_TIMEOUT_REMOVE] = {
902 /* used by timeout updates' prep() */
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000903 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300904 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700905 .needs_file = 1,
906 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700907 .pollin = 1,
Jens Axboed3656342019-12-18 09:50:26 -0700908 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300909 [IORING_OP_ASYNC_CANCEL] = {},
910 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700911 .needs_async_data = 1,
912 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700913 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300914 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700915 .needs_file = 1,
916 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700917 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700918 .needs_async_data = 1,
919 .async_size = sizeof(struct io_async_connect),
Jens Axboed3656342019-12-18 09:50:26 -0700920 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300921 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700922 .needs_file = 1,
923 },
Jens Axboe44526be2021-02-15 13:32:18 -0700924 [IORING_OP_OPENAT] = {},
925 [IORING_OP_CLOSE] = {},
926 [IORING_OP_FILES_UPDATE] = {},
927 [IORING_OP_STATX] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300928 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700929 .needs_file = 1,
930 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700931 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700932 .buffer_select = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600933 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700934 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700935 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300936 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700937 .needs_file = 1,
938 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700939 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600940 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700941 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700942 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300943 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -0700944 .needs_file = 1,
945 },
Jens Axboe44526be2021-02-15 13:32:18 -0700946 [IORING_OP_MADVISE] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300947 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700948 .needs_file = 1,
949 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700950 .pollout = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700951 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300952 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700953 .needs_file = 1,
954 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700955 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700956 .buffer_select = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700957 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300958 [IORING_OP_OPENAT2] = {
Jens Axboecebdb982020-01-08 17:59:24 -0700959 },
Jens Axboe3e4827b2020-01-08 15:18:09 -0700960 [IORING_OP_EPOLL_CTL] = {
961 .unbound_nonreg_file = 1,
Jens Axboe3e4827b2020-01-08 15:18:09 -0700962 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300963 [IORING_OP_SPLICE] = {
964 .needs_file = 1,
965 .hash_reg_file = 1,
966 .unbound_nonreg_file = 1,
Jens Axboeddf0322d2020-02-23 16:41:33 -0700967 },
968 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -0700969 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +0300970 [IORING_OP_TEE] = {
971 .needs_file = 1,
972 .hash_reg_file = 1,
973 .unbound_nonreg_file = 1,
974 },
Jens Axboe36f4fa62020-09-05 11:14:22 -0600975 [IORING_OP_SHUTDOWN] = {
976 .needs_file = 1,
977 },
Jens Axboe44526be2021-02-15 13:32:18 -0700978 [IORING_OP_RENAMEAT] = {},
979 [IORING_OP_UNLINKAT] = {},
Jens Axboed3656342019-12-18 09:50:26 -0700980};
981
Pavel Begunkov9936c7c2021-02-04 13:51:56 +0000982static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
983 struct task_struct *task,
984 struct files_struct *files);
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700985static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000986static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node);
Pavel Begunkovbc9744c2021-01-15 17:37:49 +0000987static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Pavel Begunkov1ffc5422020-12-30 21:34:15 +0000988 struct io_ring_ctx *ctx);
Pavel Begunkovf2303b12021-02-20 18:03:49 +0000989static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +0000990
Pavel Begunkov23faba32021-02-11 18:28:22 +0000991static bool io_rw_reissue(struct io_kiocb *req);
Jens Axboe78e19bb2019-11-06 15:21:34 -0700992static void io_cqring_fill_event(struct io_kiocb *req, long res);
Jackie Liuec9c02a2019-11-08 23:50:36 +0800993static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +0100994static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec40f6372020-06-25 15:39:59 -0600995static void io_double_put_req(struct io_kiocb *req);
Jens Axboec7dae4b2021-02-09 19:53:37 -0700996static void io_dismantle_req(struct io_kiocb *req);
997static void io_put_task(struct task_struct *task, int nr);
998static void io_queue_next(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -0700999static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
Jens Axboe7271ef32020-08-10 09:55:22 -06001000static void __io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001001static void io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe05f3fb32019-12-09 11:22:50 -07001002static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001003 struct io_uring_rsrc_update *ip,
Jens Axboe05f3fb32019-12-09 11:22:50 -07001004 unsigned nr_args);
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001005static void __io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01001006static struct file *io_file_get(struct io_submit_state *state,
1007 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001008static void __io_queue_sqe(struct io_kiocb *req);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001009static void io_rsrc_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -06001010
Pavel Begunkov847595d2021-02-04 13:52:06 +00001011static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
1012 struct iov_iter *iter, bool needs_lock);
Jens Axboeff6165b2020-08-13 09:47:43 -06001013static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
1014 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06001015 struct iov_iter *iter, bool force);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001016static void io_req_task_queue(struct io_kiocb *req);
Jens Axboe65453d12021-02-10 00:03:21 +00001017static void io_submit_flush_completions(struct io_comp_state *cs,
1018 struct io_ring_ctx *ctx);
Jens Axboe9a56a232019-01-09 09:06:50 -07001019
Jens Axboe2b188cc2019-01-07 10:46:33 -07001020static struct kmem_cache *req_cachep;
1021
Jens Axboe09186822020-10-13 15:01:40 -06001022static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001023
1024struct sock *io_uring_get_socket(struct file *file)
1025{
1026#if defined(CONFIG_UNIX)
1027 if (file->f_op == &io_uring_fops) {
1028 struct io_ring_ctx *ctx = file->private_data;
1029
1030 return ctx->ring_sock->sk;
1031 }
1032#endif
1033 return NULL;
1034}
1035EXPORT_SYMBOL(io_uring_get_socket);
1036
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001037#define io_for_each_link(pos, head) \
1038 for (pos = (head); pos; pos = pos->link)
1039
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001040static inline void io_clean_op(struct io_kiocb *req)
1041{
Pavel Begunkov9d5c8192021-01-24 15:08:14 +00001042 if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED))
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001043 __io_clean_op(req);
1044}
1045
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001046static inline void io_set_resource_node(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001047{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001048 struct io_ring_ctx *ctx = req->ctx;
1049
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001050 if (!req->fixed_rsrc_refs) {
1051 req->fixed_rsrc_refs = &ctx->file_data->node->refs;
1052 percpu_ref_get(req->fixed_rsrc_refs);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001053 }
1054}
1055
Pavel Begunkov08d23632020-11-06 13:00:22 +00001056static bool io_match_task(struct io_kiocb *head,
1057 struct task_struct *task,
1058 struct files_struct *files)
1059{
1060 struct io_kiocb *req;
1061
Jens Axboe84965ff2021-01-23 15:51:11 -07001062 if (task && head->task != task) {
1063 /* in terms of cancelation, always match if req task is dead */
1064 if (head->task->flags & PF_EXITING)
1065 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001066 return false;
Jens Axboe84965ff2021-01-23 15:51:11 -07001067 }
Pavel Begunkov08d23632020-11-06 13:00:22 +00001068 if (!files)
1069 return true;
1070
1071 io_for_each_link(req, head) {
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +00001072 if (req->flags & REQ_F_INFLIGHT)
Jens Axboe02a13672021-01-23 15:49:31 -07001073 return true;
Jens Axboe4379bf82021-02-15 13:40:22 -07001074 if (req->task->files == files)
Pavel Begunkov08d23632020-11-06 13:00:22 +00001075 return true;
1076 }
1077 return false;
1078}
1079
Jens Axboec40f6372020-06-25 15:39:59 -06001080static inline void req_set_fail_links(struct io_kiocb *req)
1081{
1082 if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
1083 req->flags |= REQ_F_FAIL_LINK;
1084}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001085
Jens Axboe2b188cc2019-01-07 10:46:33 -07001086static void io_ring_ctx_ref_free(struct percpu_ref *ref)
1087{
1088 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
1089
Jens Axboe0f158b42020-05-14 17:18:39 -06001090 complete(&ctx->ref_comp);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001091}
1092
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001093static inline bool io_is_timeout_noseq(struct io_kiocb *req)
1094{
1095 return !req->timeout.off;
1096}
1097
Jens Axboe2b188cc2019-01-07 10:46:33 -07001098static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
1099{
1100 struct io_ring_ctx *ctx;
Jens Axboe78076bb2019-12-04 19:56:40 -07001101 int hash_bits;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001102
1103 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1104 if (!ctx)
1105 return NULL;
1106
Jens Axboe78076bb2019-12-04 19:56:40 -07001107 /*
1108 * Use 5 bits less than the max cq entries, that should give us around
1109 * 32 entries per hash list if totally full and uniformly spread.
1110 */
1111 hash_bits = ilog2(p->cq_entries);
1112 hash_bits -= 5;
1113 if (hash_bits <= 0)
1114 hash_bits = 1;
1115 ctx->cancel_hash_bits = hash_bits;
1116 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1117 GFP_KERNEL);
1118 if (!ctx->cancel_hash)
1119 goto err;
1120 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1121
Roman Gushchin21482892019-05-07 10:01:48 -07001122 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001123 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1124 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001125
1126 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001127 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001128 INIT_LIST_HEAD(&ctx->sqd_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001129 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001130 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001131 init_completion(&ctx->ref_comp);
1132 init_completion(&ctx->sq_thread_comp);
Jens Axboe5a2e7452020-02-23 16:23:11 -07001133 idr_init(&ctx->io_buffer_idr);
Jens Axboe071698e2020-01-28 10:04:42 -07001134 idr_init(&ctx->personality_idr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001135 mutex_init(&ctx->uring_lock);
1136 init_waitqueue_head(&ctx->wait);
1137 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001138 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001139 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001140 INIT_LIST_HEAD(&ctx->timeout_list);
Jens Axboefcb323c2019-10-24 12:39:47 -06001141 spin_lock_init(&ctx->inflight_lock);
1142 INIT_LIST_HEAD(&ctx->inflight_list);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00001143 spin_lock_init(&ctx->rsrc_ref_lock);
1144 INIT_LIST_HEAD(&ctx->rsrc_ref_list);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001145 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work);
1146 init_llist_head(&ctx->rsrc_put_llist);
Jens Axboe1b4c3512021-02-10 00:03:19 +00001147 INIT_LIST_HEAD(&ctx->submit_state.comp.free_list);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001148 INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001149 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001150err:
Jens Axboe78076bb2019-12-04 19:56:40 -07001151 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001152 kfree(ctx);
1153 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001154}
1155
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001156static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001157{
Jens Axboe2bc99302020-07-09 09:43:27 -06001158 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1159 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001160
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001161 return seq != ctx->cached_cq_tail
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001162 + READ_ONCE(ctx->cached_cq_overflow);
Jens Axboe2bc99302020-07-09 09:43:27 -06001163 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001164
Bob Liu9d858b22019-11-13 18:06:25 +08001165 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001166}
1167
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001168static void io_req_track_inflight(struct io_kiocb *req)
1169{
1170 struct io_ring_ctx *ctx = req->ctx;
1171
1172 if (!(req->flags & REQ_F_INFLIGHT)) {
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001173 req->flags |= REQ_F_INFLIGHT;
1174
1175 spin_lock_irq(&ctx->inflight_lock);
1176 list_add(&req->inflight_entry, &ctx->inflight_list);
1177 spin_unlock_irq(&ctx->inflight_lock);
1178 }
1179}
1180
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001181static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001182{
Jens Axboed3656342019-12-18 09:50:26 -07001183 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001184 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe54a91f32019-09-10 09:15:04 -06001185
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001186 if (req->flags & REQ_F_FORCE_ASYNC)
1187 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1188
Jens Axboed3656342019-12-18 09:50:26 -07001189 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001190 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001191 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboed3656342019-12-18 09:50:26 -07001192 } else {
1193 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001194 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001195 }
Jens Axboe561fb042019-10-24 07:25:42 -06001196}
1197
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001198static void io_prep_async_link(struct io_kiocb *req)
1199{
1200 struct io_kiocb *cur;
1201
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001202 io_for_each_link(cur, req)
1203 io_prep_async_work(cur);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001204}
1205
Pavel Begunkovebf93662021-03-01 18:20:47 +00001206static void io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001207{
Jackie Liua197f662019-11-08 08:09:12 -07001208 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001209 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07001210 struct io_uring_task *tctx = req->task->io_uring;
Jens Axboe561fb042019-10-24 07:25:42 -06001211
Jens Axboe3bfe6102021-02-16 14:15:30 -07001212 BUG_ON(!tctx);
1213 BUG_ON(!tctx->io_wq);
Jens Axboe561fb042019-10-24 07:25:42 -06001214
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001215 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1216 &req->work, req->flags);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001217 /* init ->work of the whole link before punting */
1218 io_prep_async_link(req);
Pavel Begunkovebf93662021-03-01 18:20:47 +00001219 io_wq_enqueue(tctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001220 if (link)
1221 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001222}
1223
Jens Axboe5262f562019-09-17 12:26:57 -06001224static void io_kill_timeout(struct io_kiocb *req)
1225{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001226 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001227 int ret;
1228
Jens Axboee8c2bc12020-08-15 18:44:09 -07001229 ret = hrtimer_try_to_cancel(&io->timer);
Jens Axboe5262f562019-09-17 12:26:57 -06001230 if (ret != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001231 atomic_set(&req->ctx->cq_timeouts,
1232 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001233 list_del_init(&req->timeout.list);
Jens Axboe78e19bb2019-11-06 15:21:34 -07001234 io_cqring_fill_event(req, 0);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001235 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001236 }
1237}
1238
Jens Axboe76e1b642020-09-26 15:05:03 -06001239/*
1240 * Returns true if we found and killed one or more timeouts
1241 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00001242static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
1243 struct files_struct *files)
Jens Axboe5262f562019-09-17 12:26:57 -06001244{
1245 struct io_kiocb *req, *tmp;
Jens Axboe76e1b642020-09-26 15:05:03 -06001246 int canceled = 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001247
1248 spin_lock_irq(&ctx->completion_lock);
Jens Axboef3606e32020-09-22 08:18:24 -06001249 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00001250 if (io_match_task(req, tsk, files)) {
Jens Axboef3606e32020-09-22 08:18:24 -06001251 io_kill_timeout(req);
Jens Axboe76e1b642020-09-26 15:05:03 -06001252 canceled++;
1253 }
Jens Axboef3606e32020-09-22 08:18:24 -06001254 }
Jens Axboe5262f562019-09-17 12:26:57 -06001255 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe76e1b642020-09-26 15:05:03 -06001256 return canceled != 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001257}
1258
Pavel Begunkov04518942020-05-26 20:34:05 +03001259static void __io_queue_deferred(struct io_ring_ctx *ctx)
1260{
1261 do {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001262 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1263 struct io_defer_entry, list);
Pavel Begunkov04518942020-05-26 20:34:05 +03001264
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001265 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001266 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001267 list_del_init(&de->list);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001268 io_req_task_queue(de->req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001269 kfree(de);
Pavel Begunkov04518942020-05-26 20:34:05 +03001270 } while (!list_empty(&ctx->defer_list));
1271}
1272
Pavel Begunkov360428f2020-05-30 14:54:17 +03001273static void io_flush_timeouts(struct io_ring_ctx *ctx)
1274{
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001275 u32 seq;
1276
1277 if (list_empty(&ctx->timeout_list))
1278 return;
1279
1280 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
1281
1282 do {
1283 u32 events_needed, events_got;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001284 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001285 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001286
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001287 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001288 break;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001289
1290 /*
1291 * Since seq can easily wrap around over time, subtract
1292 * the last seq at which timeouts were flushed before comparing.
1293 * Assuming not more than 2^31-1 events have happened since,
1294 * these subtractions won't have wrapped, so we can check if
1295 * target is in [last_seq, current_seq] by comparing the two.
1296 */
1297 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
1298 events_got = seq - ctx->cq_last_tm_flush;
1299 if (events_got < events_needed)
Pavel Begunkov360428f2020-05-30 14:54:17 +03001300 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001301
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001302 list_del_init(&req->timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001303 io_kill_timeout(req);
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001304 } while (!list_empty(&ctx->timeout_list));
1305
1306 ctx->cq_last_tm_flush = seq;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001307}
1308
Jens Axboede0617e2019-04-06 21:51:27 -06001309static void io_commit_cqring(struct io_ring_ctx *ctx)
1310{
Pavel Begunkov360428f2020-05-30 14:54:17 +03001311 io_flush_timeouts(ctx);
Pavel Begunkovec30e042021-01-19 13:32:38 +00001312
1313 /* order cqe stores with ring update */
1314 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
Jens Axboede0617e2019-04-06 21:51:27 -06001315
Pavel Begunkov04518942020-05-26 20:34:05 +03001316 if (unlikely(!list_empty(&ctx->defer_list)))
1317 __io_queue_deferred(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001318}
1319
Jens Axboe90554202020-09-03 12:12:41 -06001320static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1321{
1322 struct io_rings *r = ctx->rings;
1323
1324 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries;
1325}
1326
Pavel Begunkov888aae22021-01-19 13:32:39 +00001327static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
1328{
1329 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
1330}
1331
Jens Axboe2b188cc2019-01-07 10:46:33 -07001332static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
1333{
Hristo Venev75b28af2019-08-26 17:23:46 +00001334 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001335 unsigned tail;
1336
Stefan Bühler115e12e2019-04-24 23:54:18 +02001337 /*
1338 * writes to the cq entry need to come after reading head; the
1339 * control dependency is enough as we're using WRITE_ONCE to
1340 * fill the cq entry
1341 */
Pavel Begunkov888aae22021-01-19 13:32:39 +00001342 if (__io_cqring_events(ctx) == rings->cq_ring_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001343 return NULL;
1344
Pavel Begunkov888aae22021-01-19 13:32:39 +00001345 tail = ctx->cached_cq_tail++;
Hristo Venev75b28af2019-08-26 17:23:46 +00001346 return &rings->cqes[tail & ctx->cq_mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001347}
1348
Jens Axboef2842ab2020-01-08 11:04:00 -07001349static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1350{
Jens Axboef0b493e2020-02-01 21:30:11 -07001351 if (!ctx->cq_ev_fd)
1352 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001353 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1354 return false;
Jens Axboef2842ab2020-01-08 11:04:00 -07001355 if (!ctx->eventfd_async)
1356 return true;
Jens Axboeb41e9852020-02-17 09:52:41 -07001357 return io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001358}
1359
Jens Axboeb41e9852020-02-17 09:52:41 -07001360static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001361{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001362 /* see waitqueue_active() comment */
1363 smp_mb();
1364
Jens Axboe8c838782019-03-12 15:48:16 -06001365 if (waitqueue_active(&ctx->wait))
1366 wake_up(&ctx->wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001367 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1368 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001369 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001370 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001371 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001372 wake_up_interruptible(&ctx->cq_wait);
1373 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1374 }
Jens Axboe8c838782019-03-12 15:48:16 -06001375}
1376
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001377static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
1378{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001379 /* see waitqueue_active() comment */
1380 smp_mb();
1381
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001382 if (ctx->flags & IORING_SETUP_SQPOLL) {
1383 if (waitqueue_active(&ctx->wait))
1384 wake_up(&ctx->wait);
1385 }
1386 if (io_should_trigger_evfd(ctx))
1387 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001388 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001389 wake_up_interruptible(&ctx->cq_wait);
1390 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1391 }
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001392}
1393
Jens Axboec4a2ed72019-11-21 21:01:26 -07001394/* Returns true if there are no backlogged entries after the flush */
Pavel Begunkov6c503152021-01-04 20:36:36 +00001395static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
1396 struct task_struct *tsk,
1397 struct files_struct *files)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001398{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001399 struct io_rings *rings = ctx->rings;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001400 struct io_kiocb *req, *tmp;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001401 struct io_uring_cqe *cqe;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001402 unsigned long flags;
Jens Axboeb18032b2021-01-24 16:58:56 -07001403 bool all_flushed, posted;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001404 LIST_HEAD(list);
1405
Pavel Begunkove23de152020-12-17 00:24:37 +00001406 if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries)
1407 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001408
Jens Axboeb18032b2021-01-24 16:58:56 -07001409 posted = false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001410 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboee6c8aa92020-09-28 13:10:13 -06001411 list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00001412 if (!io_match_task(req, tsk, files))
Jens Axboee6c8aa92020-09-28 13:10:13 -06001413 continue;
1414
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001415 cqe = io_get_cqring(ctx);
1416 if (!cqe && !force)
1417 break;
1418
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001419 list_move(&req->compl.list, &list);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001420 if (cqe) {
1421 WRITE_ONCE(cqe->user_data, req->user_data);
1422 WRITE_ONCE(cqe->res, req->result);
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001423 WRITE_ONCE(cqe->flags, req->compl.cflags);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001424 } else {
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001425 ctx->cached_cq_overflow++;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001426 WRITE_ONCE(ctx->rings->cq_overflow,
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001427 ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001428 }
Jens Axboeb18032b2021-01-24 16:58:56 -07001429 posted = true;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001430 }
1431
Pavel Begunkov09e88402020-12-17 00:24:38 +00001432 all_flushed = list_empty(&ctx->cq_overflow_list);
1433 if (all_flushed) {
1434 clear_bit(0, &ctx->sq_check_overflow);
1435 clear_bit(0, &ctx->cq_check_overflow);
1436 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1437 }
Pavel Begunkov46930142020-07-30 18:43:49 +03001438
Jens Axboeb18032b2021-01-24 16:58:56 -07001439 if (posted)
1440 io_commit_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001441 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeb18032b2021-01-24 16:58:56 -07001442 if (posted)
1443 io_cqring_ev_posted(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001444
1445 while (!list_empty(&list)) {
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001446 req = list_first_entry(&list, struct io_kiocb, compl.list);
1447 list_del(&req->compl.list);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001448 io_put_req(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001449 }
Jens Axboec4a2ed72019-11-21 21:01:26 -07001450
Pavel Begunkov09e88402020-12-17 00:24:38 +00001451 return all_flushed;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001452}
1453
Jens Axboeca0a2652021-03-04 17:15:48 -07001454static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
Pavel Begunkov6c503152021-01-04 20:36:36 +00001455 struct task_struct *tsk,
1456 struct files_struct *files)
1457{
Jens Axboeca0a2652021-03-04 17:15:48 -07001458 bool ret = true;
1459
Pavel Begunkov6c503152021-01-04 20:36:36 +00001460 if (test_bit(0, &ctx->cq_check_overflow)) {
1461 /* iopoll syncs against uring_lock, not completion_lock */
1462 if (ctx->flags & IORING_SETUP_IOPOLL)
1463 mutex_lock(&ctx->uring_lock);
Jens Axboeca0a2652021-03-04 17:15:48 -07001464 ret = __io_cqring_overflow_flush(ctx, force, tsk, files);
Pavel Begunkov6c503152021-01-04 20:36:36 +00001465 if (ctx->flags & IORING_SETUP_IOPOLL)
1466 mutex_unlock(&ctx->uring_lock);
1467 }
Jens Axboeca0a2652021-03-04 17:15:48 -07001468
1469 return ret;
Pavel Begunkov6c503152021-01-04 20:36:36 +00001470}
1471
Jens Axboebcda7ba2020-02-23 16:42:51 -07001472static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001473{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001474 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001475 struct io_uring_cqe *cqe;
1476
Jens Axboe78e19bb2019-11-06 15:21:34 -07001477 trace_io_uring_complete(ctx, req->user_data, res);
Jens Axboe51c3ff62019-11-03 06:52:50 -07001478
Jens Axboe2b188cc2019-01-07 10:46:33 -07001479 /*
1480 * If we can't get a cq entry, userspace overflowed the
1481 * submission (by quite a lot). Increment the overflow count in
1482 * the ring.
1483 */
1484 cqe = io_get_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001485 if (likely(cqe)) {
Jens Axboe78e19bb2019-11-06 15:21:34 -07001486 WRITE_ONCE(cqe->user_data, req->user_data);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001487 WRITE_ONCE(cqe->res, res);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001488 WRITE_ONCE(cqe->flags, cflags);
Jens Axboefdaf0832020-10-30 09:37:30 -06001489 } else if (ctx->cq_overflow_flushed ||
1490 atomic_read(&req->task->io_uring->in_idle)) {
Jens Axboe0f212202020-09-13 13:09:39 -06001491 /*
1492 * If we're in ring overflow flush mode, or in task cancel mode,
1493 * then we cannot store the request for later flushing, we need
1494 * to drop it on the floor.
1495 */
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001496 ctx->cached_cq_overflow++;
1497 WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001498 } else {
Jens Axboead3eb2c2019-12-18 17:12:20 -07001499 if (list_empty(&ctx->cq_overflow_list)) {
1500 set_bit(0, &ctx->sq_check_overflow);
1501 set_bit(0, &ctx->cq_check_overflow);
Xiaoguang Wang6d5f9042020-07-09 09:15:29 +08001502 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
Jens Axboead3eb2c2019-12-18 17:12:20 -07001503 }
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001504 io_clean_op(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001505 req->result = res;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001506 req->compl.cflags = cflags;
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001507 refcount_inc(&req->refs);
1508 list_add_tail(&req->compl.list, &ctx->cq_overflow_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001509 }
1510}
1511
Jens Axboebcda7ba2020-02-23 16:42:51 -07001512static void io_cqring_fill_event(struct io_kiocb *req, long res)
1513{
1514 __io_cqring_fill_event(req, res, 0);
1515}
1516
Jens Axboec7dae4b2021-02-09 19:53:37 -07001517static inline void io_req_complete_post(struct io_kiocb *req, long res,
1518 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001519{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001520 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001521 unsigned long flags;
1522
1523 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001524 __io_cqring_fill_event(req, res, cflags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001525 io_commit_cqring(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001526 /*
1527 * If we're the last reference to this request, add to our locked
1528 * free_list cache.
1529 */
1530 if (refcount_dec_and_test(&req->refs)) {
1531 struct io_comp_state *cs = &ctx->submit_state.comp;
1532
1533 io_dismantle_req(req);
1534 io_put_task(req->task, 1);
1535 list_add(&req->compl.list, &cs->locked_free_list);
1536 cs->locked_free_nr++;
1537 } else
1538 req = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001539 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1540
Jens Axboe8c838782019-03-12 15:48:16 -06001541 io_cqring_ev_posted(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001542 if (req) {
1543 io_queue_next(req);
1544 percpu_ref_put(&ctx->refs);
1545 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001546}
1547
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001548static void io_req_complete_state(struct io_kiocb *req, long res,
Pavel Begunkov889fca72021-02-10 00:03:09 +00001549 unsigned int cflags)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001550{
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001551 io_clean_op(req);
1552 req->result = res;
1553 req->compl.cflags = cflags;
Pavel Begunkove342c802021-01-19 13:32:47 +00001554 req->flags |= REQ_F_COMPLETE_INLINE;
Jens Axboee1e16092020-06-22 09:17:17 -06001555}
Jens Axboe2b188cc2019-01-07 10:46:33 -07001556
Pavel Begunkov889fca72021-02-10 00:03:09 +00001557static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
1558 long res, unsigned cflags)
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001559{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001560 if (issue_flags & IO_URING_F_COMPLETE_DEFER)
1561 io_req_complete_state(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001562 else
Jens Axboec7dae4b2021-02-09 19:53:37 -07001563 io_req_complete_post(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001564}
Jens Axboebcda7ba2020-02-23 16:42:51 -07001565
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001566static inline void io_req_complete(struct io_kiocb *req, long res)
Jens Axboee1e16092020-06-22 09:17:17 -06001567{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001568 __io_req_complete(req, 0, res, 0);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001569}
1570
Jens Axboec7dae4b2021-02-09 19:53:37 -07001571static bool io_flush_cached_reqs(struct io_ring_ctx *ctx)
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001572{
Jens Axboec7dae4b2021-02-09 19:53:37 -07001573 struct io_submit_state *state = &ctx->submit_state;
1574 struct io_comp_state *cs = &state->comp;
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001575 struct io_kiocb *req = NULL;
1576
Jens Axboec7dae4b2021-02-09 19:53:37 -07001577 /*
1578 * If we have more than a batch's worth of requests in our IRQ side
1579 * locked cache, grab the lock and move them over to our submission
1580 * side cache.
1581 */
1582 if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) {
1583 spin_lock_irq(&ctx->completion_lock);
1584 list_splice_init(&cs->locked_free_list, &cs->free_list);
1585 cs->locked_free_nr = 0;
1586 spin_unlock_irq(&ctx->completion_lock);
1587 }
1588
1589 while (!list_empty(&cs->free_list)) {
1590 req = list_first_entry(&cs->free_list, struct io_kiocb,
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001591 compl.list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001592 list_del(&req->compl.list);
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001593 state->reqs[state->free_reqs++] = req;
1594 if (state->free_reqs == ARRAY_SIZE(state->reqs))
1595 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001596 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001597
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001598 return req != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001599}
1600
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001601static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001602{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001603 struct io_submit_state *state = &ctx->submit_state;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001604
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001605 BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs));
Jens Axboe2b188cc2019-01-07 10:46:33 -07001606
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03001607 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03001608 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Jens Axboe2579f912019-01-09 09:10:43 -07001609 int ret;
1610
Jens Axboec7dae4b2021-02-09 19:53:37 -07001611 if (io_flush_cached_reqs(ctx))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001612 goto got_req;
1613
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001614 ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH,
1615 state->reqs);
Jens Axboefd6fab22019-03-14 16:30:06 -06001616
1617 /*
1618 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1619 * retry single alloc to be on the safe side.
1620 */
1621 if (unlikely(ret <= 0)) {
1622 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1623 if (!state->reqs[0])
Pavel Begunkov3893f392021-02-10 00:03:15 +00001624 return NULL;
Jens Axboefd6fab22019-03-14 16:30:06 -06001625 ret = 1;
1626 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03001627 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001628 }
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001629got_req:
Pavel Begunkov291b2822020-09-30 22:57:01 +03001630 state->free_reqs--;
1631 return state->reqs[state->free_reqs];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001632}
1633
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001634static inline void io_put_file(struct io_kiocb *req, struct file *file,
1635 bool fixed)
1636{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001637 if (!fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001638 fput(file);
1639}
1640
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001641static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001642{
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001643 io_clean_op(req);
Pavel Begunkov929a3af2020-02-19 00:19:09 +03001644
Jens Axboee8c2bc12020-08-15 18:44:09 -07001645 if (req->async_data)
1646 kfree(req->async_data);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001647 if (req->file)
1648 io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE));
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001649 if (req->fixed_rsrc_refs)
1650 percpu_ref_put(req->fixed_rsrc_refs);
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001651
1652 if (req->flags & REQ_F_INFLIGHT) {
1653 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001654 unsigned long flags;
1655
1656 spin_lock_irqsave(&ctx->inflight_lock, flags);
1657 list_del(&req->inflight_entry);
1658 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
1659 req->flags &= ~REQ_F_INFLIGHT;
Pavel Begunkovf85c3102021-03-01 18:20:46 +00001660 }
Pavel Begunkove6543a82020-06-28 12:52:30 +03001661}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03001662
Pavel Begunkovb23fcf42021-03-01 18:20:48 +00001663/* must to be called somewhat shortly after putting a request */
Pavel Begunkov7c660732021-01-25 11:42:21 +00001664static inline void io_put_task(struct task_struct *task, int nr)
1665{
1666 struct io_uring_task *tctx = task->io_uring;
1667
1668 percpu_counter_sub(&tctx->inflight, nr);
1669 if (unlikely(atomic_read(&tctx->in_idle)))
1670 wake_up(&tctx->wait);
1671 put_task_struct_many(task, nr);
1672}
1673
Pavel Begunkov216578e2020-10-13 09:44:00 +01001674static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03001675{
Jens Axboe51a4cc12020-08-10 10:55:56 -06001676 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001677
Pavel Begunkov216578e2020-10-13 09:44:00 +01001678 io_dismantle_req(req);
Pavel Begunkov7c660732021-01-25 11:42:21 +00001679 io_put_task(req->task, 1);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001680
Pavel Begunkov3893f392021-02-10 00:03:15 +00001681 kmem_cache_free(req_cachep, req);
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001682 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06001683}
1684
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001685static inline void io_remove_next_linked(struct io_kiocb *req)
1686{
1687 struct io_kiocb *nxt = req->link;
1688
1689 req->link = nxt->link;
1690 nxt->link = NULL;
1691}
1692
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001693static void io_kill_linked_timeout(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001694{
Jackie Liua197f662019-11-08 08:09:12 -07001695 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001696 struct io_kiocb *link;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001697 bool cancelled = false;
1698 unsigned long flags;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001699
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001700 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001701 link = req->link;
1702
Pavel Begunkov900fad42020-10-19 16:39:16 +01001703 /*
1704 * Can happen if a linked timeout fired and link had been like
1705 * req -> link t-out -> link t-out [-> ...]
1706 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001707 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
1708 struct io_timeout_data *io = link->async_data;
1709 int ret;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001710
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001711 io_remove_next_linked(req);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00001712 link->timeout.head = NULL;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001713 ret = hrtimer_try_to_cancel(&io->timer);
1714 if (ret != -1) {
1715 io_cqring_fill_event(link, -ECANCELED);
1716 io_commit_cqring(ctx);
1717 cancelled = true;
1718 }
1719 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001720 req->flags &= ~REQ_F_LINK_TIMEOUT;
Pavel Begunkov216578e2020-10-13 09:44:00 +01001721 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeab0b6452020-06-30 08:43:15 -06001722
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001723 if (cancelled) {
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001724 io_cqring_ev_posted(ctx);
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001725 io_put_req(link);
1726 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001727}
1728
Jens Axboe4d7dd462019-11-20 13:03:52 -07001729
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001730static void io_fail_links(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001731{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001732 struct io_kiocb *link, *nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07001733 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001734 unsigned long flags;
Jens Axboe9e645e112019-05-10 16:07:28 -06001735
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001736 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001737 link = req->link;
1738 req->link = NULL;
Jens Axboe9e645e112019-05-10 16:07:28 -06001739
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001740 while (link) {
1741 nxt = link->link;
1742 link->link = NULL;
1743
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02001744 trace_io_uring_fail_link(req, link);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001745 io_cqring_fill_event(link, -ECANCELED);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001746
Jens Axboe1575f212021-02-27 15:20:49 -07001747 io_put_req_deferred(link, 2);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001748 link = nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06001749 }
Jens Axboe2665abf2019-11-05 12:40:47 -07001750 io_commit_cqring(ctx);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001751 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001752
Jens Axboe2665abf2019-11-05 12:40:47 -07001753 io_cqring_ev_posted(ctx);
Jens Axboe9e645e112019-05-10 16:07:28 -06001754}
1755
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001756static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001757{
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001758 if (req->flags & REQ_F_LINK_TIMEOUT)
1759 io_kill_linked_timeout(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07001760
Jens Axboe9e645e112019-05-10 16:07:28 -06001761 /*
1762 * If LINK is set, we have dependent requests in this chain. If we
1763 * didn't fail this request, queue the first one up, moving any other
1764 * dependencies to the next request. In case of failure, fail the rest
1765 * of the chain.
1766 */
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001767 if (likely(!(req->flags & REQ_F_FAIL_LINK))) {
1768 struct io_kiocb *nxt = req->link;
1769
1770 req->link = NULL;
1771 return nxt;
1772 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03001773 io_fail_links(req);
1774 return NULL;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001775}
Jens Axboe2665abf2019-11-05 12:40:47 -07001776
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001777static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001778{
Pavel Begunkovcdbff982021-02-12 18:41:16 +00001779 if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK))))
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001780 return NULL;
1781 return __io_req_find_next(req);
1782}
1783
Pavel Begunkov2c323952021-02-28 22:04:53 +00001784static void ctx_flush_and_put(struct io_ring_ctx *ctx)
1785{
1786 if (!ctx)
1787 return;
1788 if (ctx->submit_state.comp.nr) {
1789 mutex_lock(&ctx->uring_lock);
1790 io_submit_flush_completions(&ctx->submit_state.comp, ctx);
1791 mutex_unlock(&ctx->uring_lock);
1792 }
1793 percpu_ref_put(&ctx->refs);
1794}
1795
Jens Axboe7cbf1722021-02-10 00:03:20 +00001796static bool __tctx_task_work(struct io_uring_task *tctx)
1797{
Jens Axboe65453d12021-02-10 00:03:21 +00001798 struct io_ring_ctx *ctx = NULL;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001799 struct io_wq_work_list list;
1800 struct io_wq_work_node *node;
1801
1802 if (wq_list_empty(&tctx->task_list))
1803 return false;
1804
Jens Axboe0b81e802021-02-16 10:33:53 -07001805 spin_lock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001806 list = tctx->task_list;
1807 INIT_WQ_LIST(&tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001808 spin_unlock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001809
1810 node = list.first;
1811 while (node) {
1812 struct io_wq_work_node *next = node->next;
1813 struct io_kiocb *req;
1814
1815 req = container_of(node, struct io_kiocb, io_task_work.node);
Pavel Begunkov2c323952021-02-28 22:04:53 +00001816 if (req->ctx != ctx) {
1817 ctx_flush_and_put(ctx);
1818 ctx = req->ctx;
1819 percpu_ref_get(&ctx->refs);
1820 }
1821
Jens Axboe7cbf1722021-02-10 00:03:20 +00001822 req->task_work.func(&req->task_work);
1823 node = next;
Jens Axboe65453d12021-02-10 00:03:21 +00001824 }
1825
Pavel Begunkov2c323952021-02-28 22:04:53 +00001826 ctx_flush_and_put(ctx);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001827 return list.first != NULL;
1828}
1829
1830static void tctx_task_work(struct callback_head *cb)
1831{
1832 struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work);
1833
Jens Axboe1d5f3602021-02-26 14:54:16 -07001834 clear_bit(0, &tctx->task_state);
1835
Jens Axboe7cbf1722021-02-10 00:03:20 +00001836 while (__tctx_task_work(tctx))
1837 cond_resched();
Jens Axboe7cbf1722021-02-10 00:03:20 +00001838}
1839
1840static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req,
1841 enum task_work_notify_mode notify)
1842{
1843 struct io_uring_task *tctx = tsk->io_uring;
1844 struct io_wq_work_node *node, *prev;
Jens Axboe0b81e802021-02-16 10:33:53 -07001845 unsigned long flags;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001846 int ret;
1847
1848 WARN_ON_ONCE(!tctx);
1849
Jens Axboe0b81e802021-02-16 10:33:53 -07001850 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001851 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001852 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001853
1854 /* task_work already pending, we're done */
1855 if (test_bit(0, &tctx->task_state) ||
1856 test_and_set_bit(0, &tctx->task_state))
1857 return 0;
1858
1859 if (!task_work_add(tsk, &tctx->task_work, notify))
1860 return 0;
1861
1862 /*
1863 * Slow path - we failed, find and delete work. if the work is not
1864 * in the list, it got run and we're fine.
1865 */
1866 ret = 0;
Jens Axboe0b81e802021-02-16 10:33:53 -07001867 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001868 wq_list_for_each(node, prev, &tctx->task_list) {
1869 if (&req->io_task_work.node == node) {
1870 wq_list_del(&tctx->task_list, node, prev);
1871 ret = 1;
1872 break;
1873 }
1874 }
Jens Axboe0b81e802021-02-16 10:33:53 -07001875 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001876 clear_bit(0, &tctx->task_state);
1877 return ret;
1878}
1879
Jens Axboe355fb9e2020-10-22 20:19:35 -06001880static int io_req_task_work_add(struct io_kiocb *req)
Jens Axboec2c4c832020-07-01 15:37:11 -06001881{
1882 struct task_struct *tsk = req->task;
1883 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe91989c72020-10-16 09:02:26 -06001884 enum task_work_notify_mode notify;
1885 int ret;
Jens Axboec2c4c832020-07-01 15:37:11 -06001886
Jens Axboe6200b0a2020-09-13 14:38:30 -06001887 if (tsk->flags & PF_EXITING)
1888 return -ESRCH;
1889
Jens Axboec2c4c832020-07-01 15:37:11 -06001890 /*
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06001891 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
1892 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
1893 * processing task_work. There's no reliable way to tell if TWA_RESUME
1894 * will do the job.
Jens Axboec2c4c832020-07-01 15:37:11 -06001895 */
Jens Axboe91989c72020-10-16 09:02:26 -06001896 notify = TWA_NONE;
Jens Axboe355fb9e2020-10-22 20:19:35 -06001897 if (!(ctx->flags & IORING_SETUP_SQPOLL))
Jens Axboec2c4c832020-07-01 15:37:11 -06001898 notify = TWA_SIGNAL;
1899
Jens Axboe7cbf1722021-02-10 00:03:20 +00001900 ret = io_task_work_add(tsk, req, notify);
Jens Axboec2c4c832020-07-01 15:37:11 -06001901 if (!ret)
1902 wake_up_process(tsk);
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06001903
Jens Axboec2c4c832020-07-01 15:37:11 -06001904 return ret;
1905}
1906
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001907static void io_req_task_work_add_fallback(struct io_kiocb *req,
Jens Axboe7cbf1722021-02-10 00:03:20 +00001908 task_work_func_t cb)
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001909{
Jens Axboe7c25c0d2021-02-16 07:17:00 -07001910 struct io_ring_ctx *ctx = req->ctx;
1911 struct callback_head *head;
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001912
1913 init_task_work(&req->task_work, cb);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07001914 do {
1915 head = READ_ONCE(ctx->exit_task_work);
1916 req->task_work.next = head;
1917 } while (cmpxchg(&ctx->exit_task_work, head, &req->task_work) != head);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001918}
1919
Jens Axboec40f6372020-06-25 15:39:59 -06001920static void __io_req_task_cancel(struct io_kiocb *req, int error)
1921{
1922 struct io_ring_ctx *ctx = req->ctx;
1923
1924 spin_lock_irq(&ctx->completion_lock);
1925 io_cqring_fill_event(req, error);
1926 io_commit_cqring(ctx);
1927 spin_unlock_irq(&ctx->completion_lock);
1928
1929 io_cqring_ev_posted(ctx);
1930 req_set_fail_links(req);
1931 io_double_put_req(req);
1932}
1933
1934static void io_req_task_cancel(struct callback_head *cb)
1935{
1936 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06001937 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06001938
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00001939 mutex_lock(&ctx->uring_lock);
Pavel Begunkova3df76982021-02-18 22:32:52 +00001940 __io_req_task_cancel(req, req->result);
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00001941 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06001942 percpu_ref_put(&ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06001943}
1944
1945static void __io_req_task_submit(struct io_kiocb *req)
1946{
1947 struct io_ring_ctx *ctx = req->ctx;
1948
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00001949 /* ctx stays valid until unlock, even if we drop all ours ctx->refs */
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001950 mutex_lock(&ctx->uring_lock);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00001951 if (!(current->flags & PF_EXITING) && !current->in_execve)
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001952 __io_queue_sqe(req);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001953 else
Jens Axboec40f6372020-06-25 15:39:59 -06001954 __io_req_task_cancel(req, -EFAULT);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00001955 mutex_unlock(&ctx->uring_lock);
Jens Axboe9e645e112019-05-10 16:07:28 -06001956}
1957
Jens Axboec40f6372020-06-25 15:39:59 -06001958static void io_req_task_submit(struct callback_head *cb)
1959{
1960 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
1961
1962 __io_req_task_submit(req);
1963}
1964
1965static void io_req_task_queue(struct io_kiocb *req)
1966{
Jens Axboec40f6372020-06-25 15:39:59 -06001967 int ret;
1968
Jens Axboe7cbf1722021-02-10 00:03:20 +00001969 req->task_work.func = io_req_task_submit;
Jens Axboe355fb9e2020-10-22 20:19:35 -06001970 ret = io_req_task_work_add(req);
Jens Axboec40f6372020-06-25 15:39:59 -06001971 if (unlikely(ret)) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00001972 req->result = -ECANCELED;
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00001973 percpu_ref_get(&req->ctx->refs);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001974 io_req_task_work_add_fallback(req, io_req_task_cancel);
Jens Axboec40f6372020-06-25 15:39:59 -06001975 }
Jens Axboec40f6372020-06-25 15:39:59 -06001976}
1977
Pavel Begunkova3df76982021-02-18 22:32:52 +00001978static void io_req_task_queue_fail(struct io_kiocb *req, int ret)
1979{
1980 percpu_ref_get(&req->ctx->refs);
1981 req->result = ret;
1982 req->task_work.func = io_req_task_cancel;
1983
1984 if (unlikely(io_req_task_work_add(req)))
1985 io_req_task_work_add_fallback(req, io_req_task_cancel);
1986}
1987
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001988static inline void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08001989{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03001990 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03001991
Pavel Begunkov906a8c32020-06-27 14:04:55 +03001992 if (nxt)
1993 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08001994}
1995
Jens Axboe9e645e112019-05-10 16:07:28 -06001996static void io_free_req(struct io_kiocb *req)
1997{
Pavel Begunkovc3524382020-06-28 12:52:32 +03001998 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06001999 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002000}
2001
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002002struct req_batch {
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002003 struct task_struct *task;
2004 int task_refs;
Jens Axboe1b4c3512021-02-10 00:03:19 +00002005 int ctx_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002006};
2007
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002008static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002009{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002010 rb->task_refs = 0;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002011 rb->ctx_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002012 rb->task = NULL;
2013}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002014
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002015static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2016 struct req_batch *rb)
2017{
Pavel Begunkov6e833d52021-02-11 18:28:20 +00002018 if (rb->task)
Pavel Begunkov7c660732021-01-25 11:42:21 +00002019 io_put_task(rb->task, rb->task_refs);
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002020 if (rb->ctx_refs)
2021 percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002022}
2023
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002024static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
2025 struct io_submit_state *state)
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002026{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002027 io_queue_next(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002028
Jens Axboee3bc8e92020-09-24 08:45:57 -06002029 if (req->task != rb->task) {
Pavel Begunkov7c660732021-01-25 11:42:21 +00002030 if (rb->task)
2031 io_put_task(rb->task, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002032 rb->task = req->task;
2033 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002034 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002035 rb->task_refs++;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002036 rb->ctx_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002037
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01002038 io_dismantle_req(req);
Pavel Begunkovbd759042021-02-12 03:23:50 +00002039 if (state->free_reqs != ARRAY_SIZE(state->reqs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002040 state->reqs[state->free_reqs++] = req;
Pavel Begunkovbd759042021-02-12 03:23:50 +00002041 else
2042 list_add(&req->compl.list, &state->comp.free_list);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002043}
2044
Pavel Begunkov905c1722021-02-10 00:03:14 +00002045static void io_submit_flush_completions(struct io_comp_state *cs,
2046 struct io_ring_ctx *ctx)
2047{
2048 int i, nr = cs->nr;
2049 struct io_kiocb *req;
2050 struct req_batch rb;
2051
2052 io_init_req_batch(&rb);
2053 spin_lock_irq(&ctx->completion_lock);
2054 for (i = 0; i < nr; i++) {
2055 req = cs->reqs[i];
2056 __io_cqring_fill_event(req, req->result, req->compl.cflags);
2057 }
2058 io_commit_cqring(ctx);
2059 spin_unlock_irq(&ctx->completion_lock);
2060
2061 io_cqring_ev_posted(ctx);
2062 for (i = 0; i < nr; i++) {
2063 req = cs->reqs[i];
2064
2065 /* submission and completion refs */
2066 if (refcount_sub_and_test(2, &req->refs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002067 io_req_free_batch(&rb, req, &ctx->submit_state);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002068 }
2069
2070 io_req_free_batch_finish(ctx, &rb);
2071 cs->nr = 0;
Jens Axboee65ef562019-03-12 10:16:44 -06002072}
2073
Jens Axboeba816ad2019-09-28 11:36:45 -06002074/*
2075 * Drop reference to request, return next in chain (if there is one) if this
2076 * was the last reference to this request.
2077 */
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002078static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002079{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002080 struct io_kiocb *nxt = NULL;
2081
Jens Axboe2a44f462020-02-25 13:25:41 -07002082 if (refcount_dec_and_test(&req->refs)) {
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002083 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002084 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002085 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002086 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002087}
2088
Jens Axboe2b188cc2019-01-07 10:46:33 -07002089static void io_put_req(struct io_kiocb *req)
2090{
Jens Axboedef596e2019-01-09 08:59:42 -07002091 if (refcount_dec_and_test(&req->refs))
2092 io_free_req(req);
2093}
2094
Pavel Begunkov216578e2020-10-13 09:44:00 +01002095static void io_put_req_deferred_cb(struct callback_head *cb)
2096{
2097 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2098
2099 io_free_req(req);
2100}
2101
2102static void io_free_req_deferred(struct io_kiocb *req)
2103{
2104 int ret;
2105
Jens Axboe7cbf1722021-02-10 00:03:20 +00002106 req->task_work.func = io_put_req_deferred_cb;
Jens Axboe355fb9e2020-10-22 20:19:35 -06002107 ret = io_req_task_work_add(req);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002108 if (unlikely(ret))
2109 io_req_task_work_add_fallback(req, io_put_req_deferred_cb);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002110}
2111
2112static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2113{
2114 if (refcount_sub_and_test(refs, &req->refs))
2115 io_free_req_deferred(req);
2116}
2117
Jens Axboe978db572019-11-14 22:39:04 -07002118static void io_double_put_req(struct io_kiocb *req)
2119{
2120 /* drop both submit and complete references */
2121 if (refcount_sub_and_test(2, &req->refs))
2122 io_free_req(req);
2123}
2124
Pavel Begunkov6c503152021-01-04 20:36:36 +00002125static unsigned io_cqring_events(struct io_ring_ctx *ctx)
Jens Axboea3a0e432019-08-20 11:03:11 -06002126{
2127 /* See comment at the top of this file */
2128 smp_rmb();
Pavel Begunkove23de152020-12-17 00:24:37 +00002129 return __io_cqring_events(ctx);
Jens Axboea3a0e432019-08-20 11:03:11 -06002130}
2131
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002132static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2133{
2134 struct io_rings *rings = ctx->rings;
2135
2136 /* make sure SQ entry isn't read before tail */
2137 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2138}
2139
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002140static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002141{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002142 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002143
Jens Axboebcda7ba2020-02-23 16:42:51 -07002144 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2145 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002146 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002147 kfree(kbuf);
2148 return cflags;
2149}
2150
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002151static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2152{
2153 struct io_buffer *kbuf;
2154
2155 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2156 return io_put_kbuf(req, kbuf);
2157}
2158
Jens Axboe4c6e2772020-07-01 11:29:10 -06002159static inline bool io_run_task_work(void)
2160{
Jens Axboe6200b0a2020-09-13 14:38:30 -06002161 /*
2162 * Not safe to run on exiting task, and the task_work handling will
2163 * not add work to such a task.
2164 */
2165 if (unlikely(current->flags & PF_EXITING))
2166 return false;
Jens Axboe4c6e2772020-07-01 11:29:10 -06002167 if (current->task_works) {
2168 __set_current_state(TASK_RUNNING);
2169 task_work_run();
2170 return true;
2171 }
2172
2173 return false;
2174}
2175
Jens Axboedef596e2019-01-09 08:59:42 -07002176/*
2177 * Find and free completed poll iocbs
2178 */
2179static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2180 struct list_head *done)
2181{
Jens Axboe8237e042019-12-28 10:48:22 -07002182 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002183 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002184
2185 /* order with ->result store in io_complete_rw_iopoll() */
2186 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002187
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002188 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002189 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002190 int cflags = 0;
2191
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002192 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002193 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002194
Pavel Begunkovf1613402021-02-11 18:28:21 +00002195 if (READ_ONCE(req->result) == -EAGAIN) {
2196 req->iopoll_completed = 0;
Pavel Begunkov23faba32021-02-11 18:28:22 +00002197 if (io_rw_reissue(req))
Pavel Begunkovf1613402021-02-11 18:28:21 +00002198 continue;
2199 }
2200
Jens Axboebcda7ba2020-02-23 16:42:51 -07002201 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002202 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002203
2204 __io_cqring_fill_event(req, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002205 (*nr_events)++;
2206
Pavel Begunkovc3524382020-06-28 12:52:32 +03002207 if (refcount_dec_and_test(&req->refs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002208 io_req_free_batch(&rb, req, &ctx->submit_state);
Jens Axboedef596e2019-01-09 08:59:42 -07002209 }
Jens Axboedef596e2019-01-09 08:59:42 -07002210
Jens Axboe09bb8392019-03-13 12:39:28 -06002211 io_commit_cqring(ctx);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00002212 io_cqring_ev_posted_iopoll(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002213 io_req_free_batch_finish(ctx, &rb);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002214}
2215
Jens Axboedef596e2019-01-09 08:59:42 -07002216static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2217 long min)
2218{
2219 struct io_kiocb *req, *tmp;
2220 LIST_HEAD(done);
2221 bool spin;
2222 int ret;
2223
2224 /*
2225 * Only spin for completions if we don't have multiple devices hanging
2226 * off our complete list, and we're under the requested amount.
2227 */
2228 spin = !ctx->poll_multi_file && *nr_events < min;
2229
2230 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002231 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002232 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002233
2234 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002235 * Move completed and retryable entries to our local lists.
2236 * If we find a request that requires polling, break out
2237 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002238 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002239 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002240 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002241 continue;
2242 }
2243 if (!list_empty(&done))
2244 break;
2245
2246 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2247 if (ret < 0)
2248 break;
2249
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002250 /* iopoll may have completed current req */
2251 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002252 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002253
Jens Axboedef596e2019-01-09 08:59:42 -07002254 if (ret && spin)
2255 spin = false;
2256 ret = 0;
2257 }
2258
2259 if (!list_empty(&done))
2260 io_iopoll_complete(ctx, nr_events, &done);
2261
2262 return ret;
2263}
2264
2265/*
Brian Gianforcarod195a662019-12-13 03:09:50 -08002266 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
Jens Axboedef596e2019-01-09 08:59:42 -07002267 * non-spinning poll check - we'll still enter the driver poll loop, but only
2268 * as a non-spinning completion check.
2269 */
2270static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
2271 long min)
2272{
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002273 while (!list_empty(&ctx->iopoll_list) && !need_resched()) {
Jens Axboedef596e2019-01-09 08:59:42 -07002274 int ret;
2275
2276 ret = io_do_iopoll(ctx, nr_events, min);
2277 if (ret < 0)
2278 return ret;
Pavel Begunkoveba0a4d2020-07-06 17:59:30 +03002279 if (*nr_events >= min)
Jens Axboedef596e2019-01-09 08:59:42 -07002280 return 0;
2281 }
2282
2283 return 1;
2284}
2285
2286/*
2287 * We can't just wait for polled events to come to us, we have to actively
2288 * find and complete them.
2289 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002290static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002291{
2292 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2293 return;
2294
2295 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002296 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002297 unsigned int nr_events = 0;
2298
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002299 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002300
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002301 /* let it sleep and repeat later if can't complete a request */
2302 if (nr_events == 0)
2303 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002304 /*
2305 * Ensure we allow local-to-the-cpu processing to take place,
2306 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002307 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002308 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002309 if (need_resched()) {
2310 mutex_unlock(&ctx->uring_lock);
2311 cond_resched();
2312 mutex_lock(&ctx->uring_lock);
2313 }
Jens Axboedef596e2019-01-09 08:59:42 -07002314 }
2315 mutex_unlock(&ctx->uring_lock);
2316}
2317
Pavel Begunkov7668b922020-07-07 16:36:21 +03002318static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002319{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002320 unsigned int nr_events = 0;
Jens Axboe2b2ed972019-10-25 10:06:15 -06002321 int iters = 0, ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002322
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002323 /*
2324 * We disallow the app entering submit/complete with polling, but we
2325 * still need to lock the ring to prevent racing with polled issue
2326 * that got punted to a workqueue.
2327 */
2328 mutex_lock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002329 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002330 /*
Jens Axboea3a0e432019-08-20 11:03:11 -06002331 * Don't enter poll loop if we already have events pending.
2332 * If we do, we can potentially be spinning for commands that
2333 * already triggered a CQE (eg in error).
2334 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00002335 if (test_bit(0, &ctx->cq_check_overflow))
2336 __io_cqring_overflow_flush(ctx, false, NULL, NULL);
2337 if (io_cqring_events(ctx))
Jens Axboea3a0e432019-08-20 11:03:11 -06002338 break;
2339
2340 /*
Jens Axboe500f9fb2019-08-19 12:15:59 -06002341 * If a submit got punted to a workqueue, we can have the
2342 * application entering polling for a command before it gets
2343 * issued. That app will hold the uring_lock for the duration
2344 * of the poll right here, so we need to take a breather every
2345 * now and then to ensure that the issue has a chance to add
2346 * the poll to the issued list. Otherwise we can spin here
2347 * forever, while the workqueue is stuck trying to acquire the
2348 * very same mutex.
2349 */
2350 if (!(++iters & 7)) {
2351 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002352 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002353 mutex_lock(&ctx->uring_lock);
2354 }
2355
Pavel Begunkov7668b922020-07-07 16:36:21 +03002356 ret = io_iopoll_getevents(ctx, &nr_events, min);
Jens Axboedef596e2019-01-09 08:59:42 -07002357 if (ret <= 0)
2358 break;
2359 ret = 0;
Pavel Begunkov7668b922020-07-07 16:36:21 +03002360 } while (min && !nr_events && !need_resched());
Jens Axboedef596e2019-01-09 08:59:42 -07002361
Jens Axboe500f9fb2019-08-19 12:15:59 -06002362 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002363 return ret;
2364}
2365
Jens Axboe491381ce2019-10-17 09:20:46 -06002366static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002367{
Jens Axboe491381ce2019-10-17 09:20:46 -06002368 /*
2369 * Tell lockdep we inherited freeze protection from submission
2370 * thread.
2371 */
2372 if (req->flags & REQ_F_ISREG) {
2373 struct inode *inode = file_inode(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002374
Jens Axboe491381ce2019-10-17 09:20:46 -06002375 __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002376 }
Jens Axboe491381ce2019-10-17 09:20:46 -06002377 file_end_write(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002378}
2379
Jens Axboeb63534c2020-06-04 11:28:00 -06002380#ifdef CONFIG_BLOCK
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002381static bool io_resubmit_prep(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002382{
2383 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Colin Ian King4a245472021-02-10 20:00:07 +00002384 int rw, ret;
Jens Axboeb63534c2020-06-04 11:28:00 -06002385 struct iov_iter iter;
Jens Axboeb63534c2020-06-04 11:28:00 -06002386
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002387 /* already prepared */
2388 if (req->async_data)
2389 return true;
Jens Axboeb63534c2020-06-04 11:28:00 -06002390
2391 switch (req->opcode) {
2392 case IORING_OP_READV:
2393 case IORING_OP_READ_FIXED:
2394 case IORING_OP_READ:
2395 rw = READ;
2396 break;
2397 case IORING_OP_WRITEV:
2398 case IORING_OP_WRITE_FIXED:
2399 case IORING_OP_WRITE:
2400 rw = WRITE;
2401 break;
2402 default:
2403 printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n",
2404 req->opcode);
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002405 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002406 }
2407
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002408 ret = io_import_iovec(rw, req, &iovec, &iter, false);
2409 if (ret < 0)
2410 return false;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00002411 return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false);
Jens Axboeb63534c2020-06-04 11:28:00 -06002412}
Jens Axboeb63534c2020-06-04 11:28:00 -06002413
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002414static bool io_rw_should_reissue(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002415{
Jens Axboe355afae2020-09-02 09:30:31 -06002416 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002417 struct io_ring_ctx *ctx = req->ctx;
Jens Axboeb63534c2020-06-04 11:28:00 -06002418
Jens Axboe355afae2020-09-02 09:30:31 -06002419 if (!S_ISBLK(mode) && !S_ISREG(mode))
2420 return false;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002421 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() &&
2422 !(ctx->flags & IORING_SETUP_IOPOLL)))
Jens Axboeb63534c2020-06-04 11:28:00 -06002423 return false;
Jens Axboe7c977a52021-02-23 19:17:35 -07002424 /*
2425 * If ref is dying, we might be running poll reap from the exit work.
2426 * Don't attempt to reissue from that path, just let it fail with
2427 * -EAGAIN.
2428 */
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002429 if (percpu_ref_is_dying(&ctx->refs))
2430 return false;
2431 return true;
2432}
2433#endif
2434
2435static bool io_rw_reissue(struct io_kiocb *req)
2436{
2437#ifdef CONFIG_BLOCK
2438 if (!io_rw_should_reissue(req))
Jens Axboe7c977a52021-02-23 19:17:35 -07002439 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002440
Pavel Begunkov55e6ac12021-01-08 20:57:22 +00002441 lockdep_assert_held(&req->ctx->uring_lock);
2442
Jens Axboe37d1e2e2021-02-17 21:03:43 -07002443 if (io_resubmit_prep(req)) {
Jens Axboefdee9462020-08-27 16:46:24 -06002444 refcount_inc(&req->refs);
2445 io_queue_async_work(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002446 return true;
Jens Axboefdee9462020-08-27 16:46:24 -06002447 }
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002448 req_set_fail_links(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002449#endif
2450 return false;
2451}
2452
Jens Axboea1d7c392020-06-22 11:09:46 -06002453static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002454 unsigned int issue_flags)
Jens Axboea1d7c392020-06-22 11:09:46 -06002455{
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002456 int cflags = 0;
2457
Pavel Begunkov23faba32021-02-11 18:28:22 +00002458 if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req))
2459 return;
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002460 if (res != req->result)
2461 req_set_fail_links(req);
Pavel Begunkov23faba32021-02-11 18:28:22 +00002462
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002463 if (req->rw.kiocb.ki_flags & IOCB_WRITE)
2464 kiocb_end_write(req);
2465 if (req->flags & REQ_F_BUFFER_SELECTED)
2466 cflags = io_put_rw_kbuf(req);
2467 __io_req_complete(req, issue_flags, res, cflags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002468}
2469
2470static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2471{
Jens Axboe9adbd452019-12-20 08:45:55 -07002472 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002473
Pavel Begunkov889fca72021-02-10 00:03:09 +00002474 __io_complete_rw(req, res, res2, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002475}
2476
Jens Axboedef596e2019-01-09 08:59:42 -07002477static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2478{
Jens Axboe9adbd452019-12-20 08:45:55 -07002479 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002480
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002481#ifdef CONFIG_BLOCK
2482 /* Rewind iter, if we have one. iopoll path resubmits as usual */
2483 if (res == -EAGAIN && io_rw_should_reissue(req)) {
2484 struct io_async_rw *rw = req->async_data;
2485
2486 if (rw)
2487 iov_iter_revert(&rw->iter,
2488 req->result - iov_iter_count(&rw->iter));
2489 else if (!io_resubmit_prep(req))
2490 res = -EIO;
2491 }
2492#endif
2493
Jens Axboe491381ce2019-10-17 09:20:46 -06002494 if (kiocb->ki_flags & IOCB_WRITE)
2495 kiocb_end_write(req);
Jens Axboedef596e2019-01-09 08:59:42 -07002496
Xiaoguang Wang2d7d6792020-06-16 02:06:37 +08002497 if (res != -EAGAIN && res != req->result)
Jens Axboe4e88d6e2019-12-07 20:59:47 -07002498 req_set_fail_links(req);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002499
2500 WRITE_ONCE(req->result, res);
2501 /* order with io_poll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002502 smp_wmb();
2503 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002504}
2505
2506/*
2507 * After the iocb has been issued, it's safe to be found on the poll list.
2508 * Adding the kiocb to the list AFTER submission ensures that we don't
2509 * find it from a io_iopoll_getevents() thread before the issuer is done
2510 * accessing the kiocb cookie.
2511 */
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002512static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async)
Jens Axboedef596e2019-01-09 08:59:42 -07002513{
2514 struct io_ring_ctx *ctx = req->ctx;
2515
2516 /*
2517 * Track whether we have multiple files in our lists. This will impact
2518 * how we do polling eventually, not spinning if we're on potentially
2519 * different devices.
2520 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002521 if (list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002522 ctx->poll_multi_file = false;
2523 } else if (!ctx->poll_multi_file) {
2524 struct io_kiocb *list_req;
2525
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002526 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002527 inflight_entry);
Jens Axboe9adbd452019-12-20 08:45:55 -07002528 if (list_req->file != req->file)
Jens Axboedef596e2019-01-09 08:59:42 -07002529 ctx->poll_multi_file = true;
2530 }
2531
2532 /*
2533 * For fast devices, IO may have already completed. If it has, add
2534 * it to the front so we find it first.
2535 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002536 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002537 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002538 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002539 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002540
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002541 /*
2542 * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread
2543 * task context or in io worker task context. If current task context is
2544 * sq thread, we don't need to check whether should wake up sq thread.
2545 */
2546 if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) &&
Jens Axboe534ca6d2020-09-02 13:52:19 -06002547 wq_has_sleeper(&ctx->sq_data->wait))
2548 wake_up(&ctx->sq_data->wait);
Jens Axboedef596e2019-01-09 08:59:42 -07002549}
2550
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002551static inline void io_state_file_put(struct io_submit_state *state)
2552{
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002553 if (state->file_refs) {
2554 fput_many(state->file, state->file_refs);
2555 state->file_refs = 0;
2556 }
Jens Axboe9a56a232019-01-09 09:06:50 -07002557}
2558
2559/*
2560 * Get as many references to a file as we have IOs left in this submission,
2561 * assuming most submissions are for one file, or at least that each file
2562 * has more than one submission.
2563 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002564static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002565{
2566 if (!state)
2567 return fget(fd);
2568
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002569 if (state->file_refs) {
Jens Axboe9a56a232019-01-09 09:06:50 -07002570 if (state->fd == fd) {
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002571 state->file_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002572 return state->file;
2573 }
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002574 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002575 }
2576 state->file = fget_many(fd, state->ios_left);
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002577 if (unlikely(!state->file))
Jens Axboe9a56a232019-01-09 09:06:50 -07002578 return NULL;
2579
2580 state->fd = fd;
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002581 state->file_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002582 return state->file;
2583}
2584
Jens Axboe4503b762020-06-01 10:00:27 -06002585static bool io_bdev_nowait(struct block_device *bdev)
2586{
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002587 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002588}
2589
Jens Axboe2b188cc2019-01-07 10:46:33 -07002590/*
2591 * If we tracked the file through the SCM inflight mechanism, we could support
2592 * any file. For now, just ensure that anything potentially problematic is done
2593 * inline.
2594 */
Jens Axboeaf197f52020-04-28 13:15:06 -06002595static bool io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002596{
2597 umode_t mode = file_inode(file)->i_mode;
2598
Jens Axboe4503b762020-06-01 10:00:27 -06002599 if (S_ISBLK(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002600 if (IS_ENABLED(CONFIG_BLOCK) &&
2601 io_bdev_nowait(I_BDEV(file->f_mapping->host)))
Jens Axboe4503b762020-06-01 10:00:27 -06002602 return true;
2603 return false;
2604 }
2605 if (S_ISCHR(mode) || S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002606 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002607 if (S_ISREG(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002608 if (IS_ENABLED(CONFIG_BLOCK) &&
2609 io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
Jens Axboe4503b762020-06-01 10:00:27 -06002610 file->f_op != &io_uring_fops)
2611 return true;
2612 return false;
2613 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002614
Jens Axboec5b85622020-06-09 19:23:05 -06002615 /* any ->read/write should understand O_NONBLOCK */
2616 if (file->f_flags & O_NONBLOCK)
2617 return true;
2618
Jens Axboeaf197f52020-04-28 13:15:06 -06002619 if (!(file->f_mode & FMODE_NOWAIT))
2620 return false;
2621
2622 if (rw == READ)
2623 return file->f_op->read_iter != NULL;
2624
2625 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002626}
2627
Pavel Begunkova88fc402020-09-30 22:57:53 +03002628static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002629{
Jens Axboedef596e2019-01-09 08:59:42 -07002630 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002631 struct kiocb *kiocb = &req->rw.kiocb;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002632 struct file *file = req->file;
Jens Axboe09bb8392019-03-13 12:39:28 -06002633 unsigned ioprio;
2634 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002635
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002636 if (S_ISREG(file_inode(file)->i_mode))
Jens Axboe491381ce2019-10-17 09:20:46 -06002637 req->flags |= REQ_F_ISREG;
2638
Jens Axboe2b188cc2019-01-07 10:46:33 -07002639 kiocb->ki_pos = READ_ONCE(sqe->off);
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002640 if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) {
Jens Axboeba042912019-12-25 16:33:42 -07002641 req->flags |= REQ_F_CUR_POS;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002642 kiocb->ki_pos = file->f_pos;
Jens Axboeba042912019-12-25 16:33:42 -07002643 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002644 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002645 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
2646 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
2647 if (unlikely(ret))
2648 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002649
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002650 /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */
2651 if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK))
2652 req->flags |= REQ_F_NOWAIT;
2653
Jens Axboe2b188cc2019-01-07 10:46:33 -07002654 ioprio = READ_ONCE(sqe->ioprio);
2655 if (ioprio) {
2656 ret = ioprio_check_cap(ioprio);
2657 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06002658 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002659
2660 kiocb->ki_ioprio = ioprio;
2661 } else
2662 kiocb->ki_ioprio = get_current_ioprio();
2663
Jens Axboedef596e2019-01-09 08:59:42 -07002664 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07002665 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
2666 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06002667 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002668
Jens Axboedef596e2019-01-09 08:59:42 -07002669 kiocb->ki_flags |= IOCB_HIPRI;
2670 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002671 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002672 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06002673 if (kiocb->ki_flags & IOCB_HIPRI)
2674 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07002675 kiocb->ki_complete = io_complete_rw;
2676 }
Jens Axboe9adbd452019-12-20 08:45:55 -07002677
Jens Axboe3529d8c2019-12-19 18:24:38 -07002678 req->rw.addr = READ_ONCE(sqe->addr);
2679 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002680 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002681 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002682}
2683
2684static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
2685{
2686 switch (ret) {
2687 case -EIOCBQUEUED:
2688 break;
2689 case -ERESTARTSYS:
2690 case -ERESTARTNOINTR:
2691 case -ERESTARTNOHAND:
2692 case -ERESTART_RESTARTBLOCK:
2693 /*
2694 * We can't just restart the syscall, since previously
2695 * submitted sqes may already be in progress. Just fail this
2696 * IO with EINTR.
2697 */
2698 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002699 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002700 default:
2701 kiocb->ki_complete(kiocb, ret, 0);
2702 }
2703}
2704
Jens Axboea1d7c392020-06-22 11:09:46 -06002705static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002706 unsigned int issue_flags)
Jens Axboeba816ad2019-09-28 11:36:45 -06002707{
Jens Axboeba042912019-12-25 16:33:42 -07002708 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07002709 struct io_async_rw *io = req->async_data;
Jens Axboeba042912019-12-25 16:33:42 -07002710
Jens Axboe227c0c92020-08-13 11:51:40 -06002711 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07002712 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06002713 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07002714 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002715 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07002716 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002717 }
2718
Jens Axboeba042912019-12-25 16:33:42 -07002719 if (req->flags & REQ_F_CUR_POS)
2720 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03002721 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Pavel Begunkov889fca72021-02-10 00:03:09 +00002722 __io_complete_rw(req, ret, 0, issue_flags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002723 else
2724 io_rw_done(kiocb, ret);
2725}
2726
Pavel Begunkov847595d2021-02-04 13:52:06 +00002727static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter)
Jens Axboeedafcce2019-01-09 09:16:05 -07002728{
Jens Axboe9adbd452019-12-20 08:45:55 -07002729 struct io_ring_ctx *ctx = req->ctx;
2730 size_t len = req->rw.len;
Jens Axboeedafcce2019-01-09 09:16:05 -07002731 struct io_mapped_ubuf *imu;
Pavel Begunkov4be1c612020-09-06 00:45:48 +03002732 u16 index, buf_index = req->buf_index;
Jens Axboeedafcce2019-01-09 09:16:05 -07002733 size_t offset;
2734 u64 buf_addr;
2735
Jens Axboeedafcce2019-01-09 09:16:05 -07002736 if (unlikely(buf_index >= ctx->nr_user_bufs))
2737 return -EFAULT;
Jens Axboeedafcce2019-01-09 09:16:05 -07002738 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
2739 imu = &ctx->user_bufs[index];
Jens Axboe9adbd452019-12-20 08:45:55 -07002740 buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002741
2742 /* overflow */
2743 if (buf_addr + len < buf_addr)
2744 return -EFAULT;
2745 /* not inside the mapped region */
2746 if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len)
2747 return -EFAULT;
2748
2749 /*
2750 * May not be a start of buffer, set size appropriately
2751 * and advance us to the beginning.
2752 */
2753 offset = buf_addr - imu->ubuf;
2754 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06002755
2756 if (offset) {
2757 /*
2758 * Don't use iov_iter_advance() here, as it's really slow for
2759 * using the latter parts of a big fixed buffer - it iterates
2760 * over each segment manually. We can cheat a bit here, because
2761 * we know that:
2762 *
2763 * 1) it's a BVEC iter, we set it up
2764 * 2) all bvecs are PAGE_SIZE in size, except potentially the
2765 * first and last bvec
2766 *
2767 * So just find our index, and adjust the iterator afterwards.
2768 * If the offset is within the first bvec (or the whole first
2769 * bvec, just use iov_iter_advance(). This makes it easier
2770 * since we can just skip the first segment, which may not
2771 * be PAGE_SIZE aligned.
2772 */
2773 const struct bio_vec *bvec = imu->bvec;
2774
2775 if (offset <= bvec->bv_len) {
2776 iov_iter_advance(iter, offset);
2777 } else {
2778 unsigned long seg_skip;
2779
2780 /* skip first vec */
2781 offset -= bvec->bv_len;
2782 seg_skip = 1 + (offset >> PAGE_SHIFT);
2783
2784 iter->bvec = bvec + seg_skip;
2785 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02002786 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002787 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002788 }
2789 }
2790
Pavel Begunkov847595d2021-02-04 13:52:06 +00002791 return 0;
Jens Axboeedafcce2019-01-09 09:16:05 -07002792}
2793
Jens Axboebcda7ba2020-02-23 16:42:51 -07002794static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
2795{
2796 if (needs_lock)
2797 mutex_unlock(&ctx->uring_lock);
2798}
2799
2800static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
2801{
2802 /*
2803 * "Normal" inline submissions always hold the uring_lock, since we
2804 * grab it from the system call. Same is true for the SQPOLL offload.
2805 * The only exception is when we've detached the request and issue it
2806 * from an async worker thread, grab the lock for that case.
2807 */
2808 if (needs_lock)
2809 mutex_lock(&ctx->uring_lock);
2810}
2811
2812static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
2813 int bgid, struct io_buffer *kbuf,
2814 bool needs_lock)
2815{
2816 struct io_buffer *head;
2817
2818 if (req->flags & REQ_F_BUFFER_SELECTED)
2819 return kbuf;
2820
2821 io_ring_submit_lock(req->ctx, needs_lock);
2822
2823 lockdep_assert_held(&req->ctx->uring_lock);
2824
2825 head = idr_find(&req->ctx->io_buffer_idr, bgid);
2826 if (head) {
2827 if (!list_empty(&head->list)) {
2828 kbuf = list_last_entry(&head->list, struct io_buffer,
2829 list);
2830 list_del(&kbuf->list);
2831 } else {
2832 kbuf = head;
2833 idr_remove(&req->ctx->io_buffer_idr, bgid);
2834 }
2835 if (*len > kbuf->len)
2836 *len = kbuf->len;
2837 } else {
2838 kbuf = ERR_PTR(-ENOBUFS);
2839 }
2840
2841 io_ring_submit_unlock(req->ctx, needs_lock);
2842
2843 return kbuf;
2844}
2845
Jens Axboe4d954c22020-02-27 07:31:19 -07002846static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
2847 bool needs_lock)
2848{
2849 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002850 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07002851
2852 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002853 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07002854 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
2855 if (IS_ERR(kbuf))
2856 return kbuf;
2857 req->rw.addr = (u64) (unsigned long) kbuf;
2858 req->flags |= REQ_F_BUFFER_SELECTED;
2859 return u64_to_user_ptr(kbuf->addr);
2860}
2861
2862#ifdef CONFIG_COMPAT
2863static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
2864 bool needs_lock)
2865{
2866 struct compat_iovec __user *uiov;
2867 compat_ssize_t clen;
2868 void __user *buf;
2869 ssize_t len;
2870
2871 uiov = u64_to_user_ptr(req->rw.addr);
2872 if (!access_ok(uiov, sizeof(*uiov)))
2873 return -EFAULT;
2874 if (__get_user(clen, &uiov->iov_len))
2875 return -EFAULT;
2876 if (clen < 0)
2877 return -EINVAL;
2878
2879 len = clen;
2880 buf = io_rw_buffer_select(req, &len, needs_lock);
2881 if (IS_ERR(buf))
2882 return PTR_ERR(buf);
2883 iov[0].iov_base = buf;
2884 iov[0].iov_len = (compat_size_t) len;
2885 return 0;
2886}
2887#endif
2888
2889static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2890 bool needs_lock)
2891{
2892 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
2893 void __user *buf;
2894 ssize_t len;
2895
2896 if (copy_from_user(iov, uiov, sizeof(*uiov)))
2897 return -EFAULT;
2898
2899 len = iov[0].iov_len;
2900 if (len < 0)
2901 return -EINVAL;
2902 buf = io_rw_buffer_select(req, &len, needs_lock);
2903 if (IS_ERR(buf))
2904 return PTR_ERR(buf);
2905 iov[0].iov_base = buf;
2906 iov[0].iov_len = len;
2907 return 0;
2908}
2909
2910static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2911 bool needs_lock)
2912{
Jens Axboedddb3e22020-06-04 11:27:01 -06002913 if (req->flags & REQ_F_BUFFER_SELECTED) {
2914 struct io_buffer *kbuf;
2915
2916 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2917 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
2918 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07002919 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06002920 }
Pavel Begunkovdd201662020-12-19 03:15:43 +00002921 if (req->rw.len != 1)
Jens Axboe4d954c22020-02-27 07:31:19 -07002922 return -EINVAL;
2923
2924#ifdef CONFIG_COMPAT
2925 if (req->ctx->compat)
2926 return io_compat_import(req, iov, needs_lock);
2927#endif
2928
2929 return __io_iov_buffer_select(req, iov, needs_lock);
2930}
2931
Pavel Begunkov847595d2021-02-04 13:52:06 +00002932static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
2933 struct iov_iter *iter, bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002934{
Jens Axboe9adbd452019-12-20 08:45:55 -07002935 void __user *buf = u64_to_user_ptr(req->rw.addr);
2936 size_t sqe_len = req->rw.len;
Pavel Begunkov847595d2021-02-04 13:52:06 +00002937 u8 opcode = req->opcode;
Jens Axboe4d954c22020-02-27 07:31:19 -07002938 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07002939
Pavel Begunkov7d009162019-11-25 23:14:40 +03002940 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07002941 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07002942 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07002943 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002944
Jens Axboebcda7ba2020-02-23 16:42:51 -07002945 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002946 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07002947 return -EINVAL;
2948
Jens Axboe3a6820f2019-12-22 15:19:35 -07002949 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002950 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07002951 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03002952 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07002953 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06002954 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002955 }
2956
Jens Axboe3a6820f2019-12-22 15:19:35 -07002957 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
2958 *iovec = NULL;
David Laight10fc72e2020-11-07 13:16:25 +00002959 return ret;
Jens Axboe3a6820f2019-12-22 15:19:35 -07002960 }
2961
Jens Axboe4d954c22020-02-27 07:31:19 -07002962 if (req->flags & REQ_F_BUFFER_SELECT) {
2963 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Pavel Begunkov847595d2021-02-04 13:52:06 +00002964 if (!ret)
2965 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len);
Jens Axboe4d954c22020-02-27 07:31:19 -07002966 *iovec = NULL;
2967 return ret;
2968 }
2969
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02002970 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
2971 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002972}
2973
Jens Axboe0fef9482020-08-26 10:36:20 -06002974static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
2975{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03002976 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06002977}
2978
Jens Axboe32960612019-09-23 11:05:34 -06002979/*
2980 * For files that don't have ->read_iter() and ->write_iter(), handle them
2981 * by looping over ->read() or ->write() manually.
2982 */
Jens Axboe4017eb92020-10-22 14:14:12 -06002983static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06002984{
Jens Axboe4017eb92020-10-22 14:14:12 -06002985 struct kiocb *kiocb = &req->rw.kiocb;
2986 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06002987 ssize_t ret = 0;
2988
2989 /*
2990 * Don't support polled IO through this interface, and we can't
2991 * support non-blocking either. For the latter, this just causes
2992 * the kiocb to be handled from an async context.
2993 */
2994 if (kiocb->ki_flags & IOCB_HIPRI)
2995 return -EOPNOTSUPP;
2996 if (kiocb->ki_flags & IOCB_NOWAIT)
2997 return -EAGAIN;
2998
2999 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003000 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003001 ssize_t nr;
3002
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003003 if (!iov_iter_is_bvec(iter)) {
3004 iovec = iov_iter_iovec(iter);
3005 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003006 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3007 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003008 }
3009
Jens Axboe32960612019-09-23 11:05:34 -06003010 if (rw == READ) {
3011 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003012 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003013 } else {
3014 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003015 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003016 }
3017
3018 if (nr < 0) {
3019 if (!ret)
3020 ret = nr;
3021 break;
3022 }
3023 ret += nr;
3024 if (nr != iovec.iov_len)
3025 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003026 req->rw.len -= nr;
3027 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003028 iov_iter_advance(iter, nr);
3029 }
3030
3031 return ret;
3032}
3033
Jens Axboeff6165b2020-08-13 09:47:43 -06003034static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3035 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003036{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003037 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003038
Jens Axboeff6165b2020-08-13 09:47:43 -06003039 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003040 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003041 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003042 /* can only be fixed buffers, no need to do anything */
Pavel Begunkov9c3a2052020-11-23 23:20:27 +00003043 if (iov_iter_is_bvec(iter))
Jens Axboeff6165b2020-08-13 09:47:43 -06003044 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003045 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003046 unsigned iov_off = 0;
3047
3048 rw->iter.iov = rw->fast_iov;
3049 if (iter->iov != fast_iov) {
3050 iov_off = iter->iov - fast_iov;
3051 rw->iter.iov += iov_off;
3052 }
3053 if (rw->fast_iov != fast_iov)
3054 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003055 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003056 } else {
3057 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003058 }
3059}
3060
Jens Axboee8c2bc12020-08-15 18:44:09 -07003061static inline int __io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003062{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003063 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3064 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3065 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003066}
3067
Jens Axboee8c2bc12020-08-15 18:44:09 -07003068static int io_alloc_async_data(struct io_kiocb *req)
Jens Axboef67676d2019-12-02 11:03:47 -07003069{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003070 if (!io_op_defs[req->opcode].needs_async_data)
Jens Axboed3656342019-12-18 09:50:26 -07003071 return 0;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003072
Jens Axboee8c2bc12020-08-15 18:44:09 -07003073 return __io_alloc_async_data(req);
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003074}
3075
Jens Axboeff6165b2020-08-13 09:47:43 -06003076static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3077 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003078 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003079{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003080 if (!force && !io_op_defs[req->opcode].needs_async_data)
Jens Axboe74566df2020-01-13 19:23:24 -07003081 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003082 if (!req->async_data) {
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003083 if (__io_alloc_async_data(req)) {
3084 kfree(iovec);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003085 return -ENOMEM;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003086 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003087
Jens Axboeff6165b2020-08-13 09:47:43 -06003088 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003089 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003090 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003091}
3092
Pavel Begunkov73debe62020-09-30 22:57:54 +03003093static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003094{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003095 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003096 struct iovec *iov = iorw->fast_iov;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003097 int ret;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003098
Pavel Begunkov2846c482020-11-07 13:16:27 +00003099 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003100 if (unlikely(ret < 0))
3101 return ret;
3102
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003103 iorw->bytes_done = 0;
3104 iorw->free_iovec = iov;
3105 if (iov)
3106 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003107 return 0;
3108}
3109
Pavel Begunkov73debe62020-09-30 22:57:54 +03003110static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003111{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003112 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3113 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003114 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003115}
3116
Jens Axboec1dd91d2020-08-03 16:43:59 -06003117/*
3118 * This is our waitqueue callback handler, registered through lock_page_async()
3119 * when we initially tried to do the IO with the iocb armed our waitqueue.
3120 * This gets called when the page is unlocked, and we generally expect that to
3121 * happen when the page IO is completed and the page is now uptodate. This will
3122 * queue a task_work based retry of the operation, attempting to copy the data
3123 * again. If the latter fails because the page was NOT uptodate, then we will
3124 * do a thread based blocking retry of the operation. That's the unexpected
3125 * slow path.
3126 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003127static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3128 int sync, void *arg)
3129{
3130 struct wait_page_queue *wpq;
3131 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003132 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003133
3134 wpq = container_of(wait, struct wait_page_queue, wait);
3135
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003136 if (!wake_page_match(wpq, key))
3137 return 0;
3138
Hao Xuc8d317a2020-09-29 20:00:45 +08003139 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003140 list_del_init(&wait->entry);
3141
Jens Axboebcf5a062020-05-22 09:24:42 -06003142 /* submit ref gets dropped, acquire a new one */
3143 refcount_inc(&req->refs);
Pavel Begunkov921b9052021-02-12 03:23:53 +00003144 io_req_task_queue(req);
Jens Axboebcf5a062020-05-22 09:24:42 -06003145 return 1;
3146}
3147
Jens Axboec1dd91d2020-08-03 16:43:59 -06003148/*
3149 * This controls whether a given IO request should be armed for async page
3150 * based retry. If we return false here, the request is handed to the async
3151 * worker threads for retry. If we're doing buffered reads on a regular file,
3152 * we prepare a private wait_page_queue entry and retry the operation. This
3153 * will either succeed because the page is now uptodate and unlocked, or it
3154 * will register a callback when the page is unlocked at IO completion. Through
3155 * that callback, io_uring uses task_work to setup a retry of the operation.
3156 * That retry will attempt the buffered read again. The retry will generally
3157 * succeed, or in rare cases where it fails, we then fall back to using the
3158 * async worker threads for a blocking retry.
3159 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003160static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003161{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003162 struct io_async_rw *rw = req->async_data;
3163 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003164 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003165
3166 /* never retry for NOWAIT, we just complete with -EAGAIN */
3167 if (req->flags & REQ_F_NOWAIT)
3168 return false;
3169
Jens Axboe227c0c92020-08-13 11:51:40 -06003170 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003171 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003172 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003173
Jens Axboebcf5a062020-05-22 09:24:42 -06003174 /*
3175 * just use poll if we can, and don't attempt if the fs doesn't
3176 * support callback based unlocks
3177 */
3178 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3179 return false;
3180
Jens Axboe3b2a4432020-08-16 10:58:43 -07003181 wait->wait.func = io_async_buf_func;
3182 wait->wait.private = req;
3183 wait->wait.flags = 0;
3184 INIT_LIST_HEAD(&wait->wait.entry);
3185 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003186 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003187 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003188 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003189}
3190
3191static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
3192{
3193 if (req->file->f_op->read_iter)
3194 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003195 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003196 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003197 else
3198 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003199}
3200
Pavel Begunkov889fca72021-02-10 00:03:09 +00003201static int io_read(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003202{
3203 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003204 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003205 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003206 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003207 ssize_t io_size, ret, ret2;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003208 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003209
Pavel Begunkov2846c482020-11-07 13:16:27 +00003210 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003211 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003212 iovec = NULL;
3213 } else {
3214 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
3215 if (ret < 0)
3216 return ret;
3217 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003218 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003219 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003220
Jens Axboefd6c2e42019-12-18 12:19:41 -07003221 /* Ensure we clear previously set non-block flag */
3222 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003223 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003224 else
3225 kiocb->ki_flags |= IOCB_NOWAIT;
3226
Pavel Begunkov24c74672020-06-21 13:09:51 +03003227 /* If the file doesn't support async, just async punt */
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003228 if (force_nonblock && !io_file_supports_async(req->file, READ)) {
3229 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003230 return ret ?: -EAGAIN;
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003231 }
Jens Axboe9e645e112019-05-10 16:07:28 -06003232
Pavel Begunkov632546c2020-11-07 13:16:26 +00003233 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003234 if (unlikely(ret)) {
3235 kfree(iovec);
3236 return ret;
3237 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003238
Jens Axboe227c0c92020-08-13 11:51:40 -06003239 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003240
Pavel Begunkov57cd6572021-02-01 18:59:56 +00003241 if (ret == -EIOCBQUEUED) {
Jens Axboe3e6a0d32021-03-01 13:56:00 -07003242 if (req->async_data)
3243 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003244 goto out_free;
Jens Axboe227c0c92020-08-13 11:51:40 -06003245 } else if (ret == -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003246 /* IOPOLL retry should happen for io-wq threads */
3247 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003248 goto done;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003249 /* no retry on NONBLOCK nor RWF_NOWAIT */
3250 if (req->flags & REQ_F_NOWAIT)
Jens Axboe355afae2020-09-02 09:30:31 -06003251 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003252 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003253 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003254 ret = 0;
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003255 } else if (ret <= 0 || ret == io_size || !force_nonblock ||
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003256 (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) {
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003257 /* read all, failed, already did sync or don't want to retry */
Jens Axboe00d23d52020-08-25 12:59:22 -06003258 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003259 }
3260
Jens Axboe227c0c92020-08-13 11:51:40 -06003261 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003262 if (ret2)
3263 return ret2;
3264
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003265 iovec = NULL;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003266 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003267 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003268 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003269
Pavel Begunkovb23df912021-02-04 13:52:04 +00003270 do {
3271 io_size -= ret;
3272 rw->bytes_done += ret;
3273 /* if we can retry, do so with the callbacks armed */
3274 if (!io_rw_should_retry(req)) {
3275 kiocb->ki_flags &= ~IOCB_WAITQ;
3276 return -EAGAIN;
3277 }
3278
3279 /*
3280 * Now retry read with the IOCB_WAITQ parts set in the iocb. If
3281 * we get -EIOCBQUEUED, then we'll get a notification when the
3282 * desired page gets unlocked. We can also get a partial read
3283 * here, and if we do, then just retry at the new offset.
3284 */
3285 ret = io_iter_do_read(req, iter);
3286 if (ret == -EIOCBQUEUED)
3287 return 0;
Jens Axboe227c0c92020-08-13 11:51:40 -06003288 /* we got some bytes, but not all. retry. */
Pavel Begunkovb23df912021-02-04 13:52:04 +00003289 } while (ret > 0 && ret < io_size);
Jens Axboe227c0c92020-08-13 11:51:40 -06003290done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003291 kiocb_done(kiocb, ret, issue_flags);
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003292out_free:
3293 /* it's faster to check here then delegate to kfree */
3294 if (iovec)
3295 kfree(iovec);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003296 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003297}
3298
Pavel Begunkov73debe62020-09-30 22:57:54 +03003299static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003300{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003301 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3302 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003303 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003304}
3305
Pavel Begunkov889fca72021-02-10 00:03:09 +00003306static int io_write(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003307{
3308 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003309 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003310 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003311 struct io_async_rw *rw = req->async_data;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003312 ssize_t ret, ret2, io_size;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003313 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003314
Pavel Begunkov2846c482020-11-07 13:16:27 +00003315 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003316 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003317 iovec = NULL;
3318 } else {
3319 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
3320 if (ret < 0)
3321 return ret;
3322 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003323 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003324 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003325
Jens Axboefd6c2e42019-12-18 12:19:41 -07003326 /* Ensure we clear previously set non-block flag */
3327 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003328 kiocb->ki_flags &= ~IOCB_NOWAIT;
3329 else
3330 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003331
Pavel Begunkov24c74672020-06-21 13:09:51 +03003332 /* If the file doesn't support async, just async punt */
Jens Axboeaf197f52020-04-28 13:15:06 -06003333 if (force_nonblock && !io_file_supports_async(req->file, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003334 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003335
Jens Axboe10d59342019-12-09 20:16:22 -07003336 /* file path doesn't support NOWAIT for non-direct_IO */
3337 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3338 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003339 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003340
Pavel Begunkov632546c2020-11-07 13:16:26 +00003341 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003342 if (unlikely(ret))
3343 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003344
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003345 /*
3346 * Open-code file_start_write here to grab freeze protection,
3347 * which will be released by another thread in
3348 * io_complete_rw(). Fool lockdep by telling it the lock got
3349 * released so that it doesn't complain about the held lock when
3350 * we return to userspace.
3351 */
3352 if (req->flags & REQ_F_ISREG) {
Darrick J. Wong8a3c84b2020-11-10 16:50:21 -08003353 sb_start_write(file_inode(req->file)->i_sb);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003354 __sb_writers_release(file_inode(req->file)->i_sb,
3355 SB_FREEZE_WRITE);
3356 }
3357 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003358
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003359 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003360 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003361 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003362 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003363 else
3364 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003365
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003366 /*
3367 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3368 * retry them without IOCB_NOWAIT.
3369 */
3370 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3371 ret2 = -EAGAIN;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003372 /* no retry on NONBLOCK nor RWF_NOWAIT */
3373 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT))
Jens Axboe355afae2020-09-02 09:30:31 -06003374 goto done;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07003375 if (ret2 == -EIOCBQUEUED && req->async_data)
3376 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003377 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003378 /* IOPOLL retry should happen for io-wq threads */
3379 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3380 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003381done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003382 kiocb_done(kiocb, ret2, issue_flags);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003383 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003384copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003385 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003386 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003387 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003388 return ret ?: -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003389 }
Jens Axboe31b51512019-01-18 22:56:34 -07003390out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003391 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003392 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003393 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003394 return ret;
3395}
3396
Jens Axboe80a261f2020-09-28 14:23:58 -06003397static int io_renameat_prep(struct io_kiocb *req,
3398 const struct io_uring_sqe *sqe)
3399{
3400 struct io_rename *ren = &req->rename;
3401 const char __user *oldf, *newf;
3402
3403 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3404 return -EBADF;
3405
3406 ren->old_dfd = READ_ONCE(sqe->fd);
3407 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
3408 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3409 ren->new_dfd = READ_ONCE(sqe->len);
3410 ren->flags = READ_ONCE(sqe->rename_flags);
3411
3412 ren->oldpath = getname(oldf);
3413 if (IS_ERR(ren->oldpath))
3414 return PTR_ERR(ren->oldpath);
3415
3416 ren->newpath = getname(newf);
3417 if (IS_ERR(ren->newpath)) {
3418 putname(ren->oldpath);
3419 return PTR_ERR(ren->newpath);
3420 }
3421
3422 req->flags |= REQ_F_NEED_CLEANUP;
3423 return 0;
3424}
3425
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003426static int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe80a261f2020-09-28 14:23:58 -06003427{
3428 struct io_rename *ren = &req->rename;
3429 int ret;
3430
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003431 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe80a261f2020-09-28 14:23:58 -06003432 return -EAGAIN;
3433
3434 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
3435 ren->newpath, ren->flags);
3436
3437 req->flags &= ~REQ_F_NEED_CLEANUP;
3438 if (ret < 0)
3439 req_set_fail_links(req);
3440 io_req_complete(req, ret);
3441 return 0;
3442}
3443
Jens Axboe14a11432020-09-28 14:27:37 -06003444static int io_unlinkat_prep(struct io_kiocb *req,
3445 const struct io_uring_sqe *sqe)
3446{
3447 struct io_unlink *un = &req->unlink;
3448 const char __user *fname;
3449
3450 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3451 return -EBADF;
3452
3453 un->dfd = READ_ONCE(sqe->fd);
3454
3455 un->flags = READ_ONCE(sqe->unlink_flags);
3456 if (un->flags & ~AT_REMOVEDIR)
3457 return -EINVAL;
3458
3459 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
3460 un->filename = getname(fname);
3461 if (IS_ERR(un->filename))
3462 return PTR_ERR(un->filename);
3463
3464 req->flags |= REQ_F_NEED_CLEANUP;
3465 return 0;
3466}
3467
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003468static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe14a11432020-09-28 14:27:37 -06003469{
3470 struct io_unlink *un = &req->unlink;
3471 int ret;
3472
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003473 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe14a11432020-09-28 14:27:37 -06003474 return -EAGAIN;
3475
3476 if (un->flags & AT_REMOVEDIR)
3477 ret = do_rmdir(un->dfd, un->filename);
3478 else
3479 ret = do_unlinkat(un->dfd, un->filename);
3480
3481 req->flags &= ~REQ_F_NEED_CLEANUP;
3482 if (ret < 0)
3483 req_set_fail_links(req);
3484 io_req_complete(req, ret);
3485 return 0;
3486}
3487
Jens Axboe36f4fa62020-09-05 11:14:22 -06003488static int io_shutdown_prep(struct io_kiocb *req,
3489 const struct io_uring_sqe *sqe)
3490{
3491#if defined(CONFIG_NET)
3492 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3493 return -EINVAL;
3494 if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags ||
3495 sqe->buf_index)
3496 return -EINVAL;
3497
3498 req->shutdown.how = READ_ONCE(sqe->len);
3499 return 0;
3500#else
3501 return -EOPNOTSUPP;
3502#endif
3503}
3504
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003505static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003506{
3507#if defined(CONFIG_NET)
3508 struct socket *sock;
3509 int ret;
3510
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003511 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003512 return -EAGAIN;
3513
Linus Torvalds48aba792020-12-16 12:44:05 -08003514 sock = sock_from_file(req->file);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003515 if (unlikely(!sock))
Linus Torvalds48aba792020-12-16 12:44:05 -08003516 return -ENOTSOCK;
Jens Axboe36f4fa62020-09-05 11:14:22 -06003517
3518 ret = __sys_shutdown_sock(sock, req->shutdown.how);
Jens Axboea1464682020-12-14 20:57:27 -07003519 if (ret < 0)
3520 req_set_fail_links(req);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003521 io_req_complete(req, ret);
3522 return 0;
3523#else
3524 return -EOPNOTSUPP;
3525#endif
3526}
3527
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003528static int __io_splice_prep(struct io_kiocb *req,
3529 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003530{
3531 struct io_splice* sp = &req->splice;
3532 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003533
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003534 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3535 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003536
3537 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003538 sp->len = READ_ONCE(sqe->len);
3539 sp->flags = READ_ONCE(sqe->splice_flags);
3540
3541 if (unlikely(sp->flags & ~valid_flags))
3542 return -EINVAL;
3543
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003544 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3545 (sp->flags & SPLICE_F_FD_IN_FIXED));
3546 if (!sp->file_in)
3547 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003548 req->flags |= REQ_F_NEED_CLEANUP;
3549
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003550 if (!S_ISREG(file_inode(sp->file_in)->i_mode)) {
3551 /*
3552 * Splice operation will be punted aync, and here need to
3553 * modify io_wq_work.flags, so initialize io_wq_work firstly.
3554 */
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003555 req->work.flags |= IO_WQ_WORK_UNBOUND;
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003556 }
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003557
3558 return 0;
3559}
3560
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003561static int io_tee_prep(struct io_kiocb *req,
3562 const struct io_uring_sqe *sqe)
3563{
3564 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3565 return -EINVAL;
3566 return __io_splice_prep(req, sqe);
3567}
3568
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003569static int io_tee(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003570{
3571 struct io_splice *sp = &req->splice;
3572 struct file *in = sp->file_in;
3573 struct file *out = sp->file_out;
3574 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3575 long ret = 0;
3576
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003577 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003578 return -EAGAIN;
3579 if (sp->len)
3580 ret = do_tee(in, out, sp->len, flags);
3581
3582 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3583 req->flags &= ~REQ_F_NEED_CLEANUP;
3584
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003585 if (ret != sp->len)
3586 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003587 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003588 return 0;
3589}
3590
3591static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3592{
3593 struct io_splice* sp = &req->splice;
3594
3595 sp->off_in = READ_ONCE(sqe->splice_off_in);
3596 sp->off_out = READ_ONCE(sqe->off);
3597 return __io_splice_prep(req, sqe);
3598}
3599
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003600static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003601{
3602 struct io_splice *sp = &req->splice;
3603 struct file *in = sp->file_in;
3604 struct file *out = sp->file_out;
3605 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3606 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003607 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003608
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003609 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003610 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003611
3612 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3613 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003614
Jens Axboe948a7742020-05-17 14:21:38 -06003615 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003616 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003617
3618 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3619 req->flags &= ~REQ_F_NEED_CLEANUP;
3620
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003621 if (ret != sp->len)
3622 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003623 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003624 return 0;
3625}
3626
Jens Axboe2b188cc2019-01-07 10:46:33 -07003627/*
3628 * IORING_OP_NOP just posts a completion event, nothing else.
3629 */
Pavel Begunkov889fca72021-02-10 00:03:09 +00003630static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003631{
3632 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003633
Jens Axboedef596e2019-01-09 08:59:42 -07003634 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
3635 return -EINVAL;
3636
Pavel Begunkov889fca72021-02-10 00:03:09 +00003637 __io_req_complete(req, issue_flags, 0, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003638 return 0;
3639}
3640
Pavel Begunkov1155c762021-02-18 18:29:38 +00003641static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003642{
Jens Axboe6b063142019-01-10 22:13:58 -07003643 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003644
Jens Axboe09bb8392019-03-13 12:39:28 -06003645 if (!req->file)
3646 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003647
Jens Axboe6b063142019-01-10 22:13:58 -07003648 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07003649 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07003650 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003651 return -EINVAL;
3652
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003653 req->sync.flags = READ_ONCE(sqe->fsync_flags);
3654 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
3655 return -EINVAL;
3656
3657 req->sync.off = READ_ONCE(sqe->off);
3658 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003659 return 0;
3660}
3661
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003662static int io_fsync(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe78912932020-01-14 22:09:06 -07003663{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003664 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003665 int ret;
3666
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003667 /* fsync always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003668 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003669 return -EAGAIN;
3670
Jens Axboe9adbd452019-12-20 08:45:55 -07003671 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003672 end > 0 ? end : LLONG_MAX,
3673 req->sync.flags & IORING_FSYNC_DATASYNC);
3674 if (ret < 0)
3675 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003676 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003677 return 0;
3678}
3679
Jens Axboed63d1b52019-12-10 10:38:56 -07003680static int io_fallocate_prep(struct io_kiocb *req,
3681 const struct io_uring_sqe *sqe)
3682{
3683 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
3684 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003685 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3686 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07003687
3688 req->sync.off = READ_ONCE(sqe->off);
3689 req->sync.len = READ_ONCE(sqe->addr);
3690 req->sync.mode = READ_ONCE(sqe->len);
3691 return 0;
3692}
3693
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003694static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboed63d1b52019-12-10 10:38:56 -07003695{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003696 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07003697
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003698 /* fallocate always requiring blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003699 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003700 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003701 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
3702 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003703 if (ret < 0)
3704 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003705 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07003706 return 0;
3707}
3708
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003709static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003710{
Jens Axboef8748882020-01-08 17:47:02 -07003711 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003712 int ret;
3713
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003714 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07003715 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003716 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07003717 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003718
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003719 /* open.how should be already initialised */
3720 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06003721 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003722
Pavel Begunkov25e72d12020-06-03 18:03:23 +03003723 req->open.dfd = READ_ONCE(sqe->fd);
3724 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07003725 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003726 if (IS_ERR(req->open.filename)) {
3727 ret = PTR_ERR(req->open.filename);
3728 req->open.filename = NULL;
3729 return ret;
3730 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06003731 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003732 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003733 return 0;
3734}
3735
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003736static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3737{
3738 u64 flags, mode;
3739
Jens Axboe14587a462020-09-05 11:36:08 -06003740 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003741 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003742 mode = READ_ONCE(sqe->len);
3743 flags = READ_ONCE(sqe->open_flags);
3744 req->open.how = build_open_how(flags, mode);
3745 return __io_openat_prep(req, sqe);
3746}
3747
Jens Axboecebdb982020-01-08 17:59:24 -07003748static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3749{
3750 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07003751 size_t len;
3752 int ret;
3753
Jens Axboe14587a462020-09-05 11:36:08 -06003754 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003755 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07003756 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3757 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07003758 if (len < OPEN_HOW_SIZE_VER0)
3759 return -EINVAL;
3760
3761 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
3762 len);
3763 if (ret)
3764 return ret;
3765
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003766 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07003767}
3768
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003769static int io_openat2(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003770{
3771 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003772 struct file *file;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003773 bool nonblock_set;
3774 bool resolve_nonblock;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003775 int ret;
3776
Jens Axboecebdb982020-01-08 17:59:24 -07003777 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003778 if (ret)
3779 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003780 nonblock_set = op.open_flag & O_NONBLOCK;
3781 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003782 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003783 /*
3784 * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open,
3785 * it'll always -EAGAIN
3786 */
3787 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE))
3788 return -EAGAIN;
3789 op.lookup_flags |= LOOKUP_CACHED;
3790 op.open_flag |= O_NONBLOCK;
3791 }
Jens Axboe15b71ab2019-12-11 11:20:36 -07003792
Jens Axboe4022e7a2020-03-19 19:23:18 -06003793 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003794 if (ret < 0)
3795 goto err;
3796
3797 file = do_filp_open(req->open.dfd, req->open.filename, &op);
Jens Axboe3a81fd02020-12-10 12:25:36 -07003798 /* only retry if RESOLVE_CACHED wasn't already set by application */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003799 if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) &&
3800 file == ERR_PTR(-EAGAIN)) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003801 /*
3802 * We could hang on to this 'fd', but seems like marginal
3803 * gain for something that is now known to be a slower path.
3804 * So just put it, and we'll get a new one when we retry.
3805 */
3806 put_unused_fd(ret);
3807 return -EAGAIN;
3808 }
3809
Jens Axboe15b71ab2019-12-11 11:20:36 -07003810 if (IS_ERR(file)) {
3811 put_unused_fd(ret);
3812 ret = PTR_ERR(file);
3813 } else {
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003814 if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set)
Jens Axboe3a81fd02020-12-10 12:25:36 -07003815 file->f_flags &= ~O_NONBLOCK;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003816 fsnotify_open(file);
3817 fd_install(ret, file);
3818 }
3819err:
3820 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003821 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003822 if (ret < 0)
3823 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003824 io_req_complete(req, ret);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003825 return 0;
3826}
3827
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003828static int io_openat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboecebdb982020-01-08 17:59:24 -07003829{
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003830 return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK);
Jens Axboecebdb982020-01-08 17:59:24 -07003831}
3832
Jens Axboe067524e2020-03-02 16:32:28 -07003833static int io_remove_buffers_prep(struct io_kiocb *req,
3834 const struct io_uring_sqe *sqe)
3835{
3836 struct io_provide_buf *p = &req->pbuf;
3837 u64 tmp;
3838
3839 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
3840 return -EINVAL;
3841
3842 tmp = READ_ONCE(sqe->fd);
3843 if (!tmp || tmp > USHRT_MAX)
3844 return -EINVAL;
3845
3846 memset(p, 0, sizeof(*p));
3847 p->nbufs = tmp;
3848 p->bgid = READ_ONCE(sqe->buf_group);
3849 return 0;
3850}
3851
3852static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
3853 int bgid, unsigned nbufs)
3854{
3855 unsigned i = 0;
3856
3857 /* shouldn't happen */
3858 if (!nbufs)
3859 return 0;
3860
3861 /* the head kbuf is the list itself */
3862 while (!list_empty(&buf->list)) {
3863 struct io_buffer *nxt;
3864
3865 nxt = list_first_entry(&buf->list, struct io_buffer, list);
3866 list_del(&nxt->list);
3867 kfree(nxt);
3868 if (++i == nbufs)
3869 return i;
3870 }
3871 i++;
3872 kfree(buf);
3873 idr_remove(&ctx->io_buffer_idr, bgid);
3874
3875 return i;
3876}
3877
Pavel Begunkov889fca72021-02-10 00:03:09 +00003878static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe067524e2020-03-02 16:32:28 -07003879{
3880 struct io_provide_buf *p = &req->pbuf;
3881 struct io_ring_ctx *ctx = req->ctx;
3882 struct io_buffer *head;
3883 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003884 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe067524e2020-03-02 16:32:28 -07003885
3886 io_ring_submit_lock(ctx, !force_nonblock);
3887
3888 lockdep_assert_held(&ctx->uring_lock);
3889
3890 ret = -ENOENT;
3891 head = idr_find(&ctx->io_buffer_idr, p->bgid);
3892 if (head)
3893 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
Jens Axboe067524e2020-03-02 16:32:28 -07003894 if (ret < 0)
3895 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003896
3897 /* need to hold the lock to complete IOPOLL requests */
3898 if (ctx->flags & IORING_SETUP_IOPOLL) {
Pavel Begunkov889fca72021-02-10 00:03:09 +00003899 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003900 io_ring_submit_unlock(ctx, !force_nonblock);
3901 } else {
3902 io_ring_submit_unlock(ctx, !force_nonblock);
Pavel Begunkov889fca72021-02-10 00:03:09 +00003903 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003904 }
Jens Axboe067524e2020-03-02 16:32:28 -07003905 return 0;
3906}
3907
Jens Axboeddf0322d2020-02-23 16:41:33 -07003908static int io_provide_buffers_prep(struct io_kiocb *req,
3909 const struct io_uring_sqe *sqe)
3910{
3911 struct io_provide_buf *p = &req->pbuf;
3912 u64 tmp;
3913
3914 if (sqe->ioprio || sqe->rw_flags)
3915 return -EINVAL;
3916
3917 tmp = READ_ONCE(sqe->fd);
3918 if (!tmp || tmp > USHRT_MAX)
3919 return -E2BIG;
3920 p->nbufs = tmp;
3921 p->addr = READ_ONCE(sqe->addr);
3922 p->len = READ_ONCE(sqe->len);
3923
Bijan Mottahedehefe68c12020-06-04 18:01:52 -07003924 if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs)))
Jens Axboeddf0322d2020-02-23 16:41:33 -07003925 return -EFAULT;
3926
3927 p->bgid = READ_ONCE(sqe->buf_group);
3928 tmp = READ_ONCE(sqe->off);
3929 if (tmp > USHRT_MAX)
3930 return -E2BIG;
3931 p->bid = tmp;
3932 return 0;
3933}
3934
3935static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
3936{
3937 struct io_buffer *buf;
3938 u64 addr = pbuf->addr;
3939 int i, bid = pbuf->bid;
3940
3941 for (i = 0; i < pbuf->nbufs; i++) {
3942 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
3943 if (!buf)
3944 break;
3945
3946 buf->addr = addr;
3947 buf->len = pbuf->len;
3948 buf->bid = bid;
3949 addr += pbuf->len;
3950 bid++;
3951 if (!*head) {
3952 INIT_LIST_HEAD(&buf->list);
3953 *head = buf;
3954 } else {
3955 list_add_tail(&buf->list, &(*head)->list);
3956 }
3957 }
3958
3959 return i ? i : -ENOMEM;
3960}
3961
Pavel Begunkov889fca72021-02-10 00:03:09 +00003962static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeddf0322d2020-02-23 16:41:33 -07003963{
3964 struct io_provide_buf *p = &req->pbuf;
3965 struct io_ring_ctx *ctx = req->ctx;
3966 struct io_buffer *head, *list;
3967 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003968 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboeddf0322d2020-02-23 16:41:33 -07003969
3970 io_ring_submit_lock(ctx, !force_nonblock);
3971
3972 lockdep_assert_held(&ctx->uring_lock);
3973
3974 list = head = idr_find(&ctx->io_buffer_idr, p->bgid);
3975
3976 ret = io_add_buffers(p, &head);
3977 if (ret < 0)
3978 goto out;
3979
3980 if (!list) {
3981 ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1,
3982 GFP_KERNEL);
3983 if (ret < 0) {
Jens Axboe067524e2020-03-02 16:32:28 -07003984 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07003985 goto out;
3986 }
3987 }
3988out:
Jens Axboeddf0322d2020-02-23 16:41:33 -07003989 if (ret < 0)
3990 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003991
3992 /* need to hold the lock to complete IOPOLL requests */
3993 if (ctx->flags & IORING_SETUP_IOPOLL) {
Pavel Begunkov889fca72021-02-10 00:03:09 +00003994 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003995 io_ring_submit_unlock(ctx, !force_nonblock);
3996 } else {
3997 io_ring_submit_unlock(ctx, !force_nonblock);
Pavel Begunkov889fca72021-02-10 00:03:09 +00003998 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003999 }
Jens Axboeddf0322d2020-02-23 16:41:33 -07004000 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004001}
4002
Jens Axboe3e4827b2020-01-08 15:18:09 -07004003static int io_epoll_ctl_prep(struct io_kiocb *req,
4004 const struct io_uring_sqe *sqe)
4005{
4006#if defined(CONFIG_EPOLL)
4007 if (sqe->ioprio || sqe->buf_index)
4008 return -EINVAL;
Jens Axboe6ca56f82020-09-18 16:51:19 -06004009 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004010 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004011
4012 req->epoll.epfd = READ_ONCE(sqe->fd);
4013 req->epoll.op = READ_ONCE(sqe->len);
4014 req->epoll.fd = READ_ONCE(sqe->off);
4015
4016 if (ep_op_has_event(req->epoll.op)) {
4017 struct epoll_event __user *ev;
4018
4019 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4020 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4021 return -EFAULT;
4022 }
4023
4024 return 0;
4025#else
4026 return -EOPNOTSUPP;
4027#endif
4028}
4029
Pavel Begunkov889fca72021-02-10 00:03:09 +00004030static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004031{
4032#if defined(CONFIG_EPOLL)
4033 struct io_epoll *ie = &req->epoll;
4034 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004035 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004036
4037 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4038 if (force_nonblock && ret == -EAGAIN)
4039 return -EAGAIN;
4040
4041 if (ret < 0)
4042 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004043 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004044 return 0;
4045#else
4046 return -EOPNOTSUPP;
4047#endif
4048}
4049
Jens Axboec1ca7572019-12-25 22:18:28 -07004050static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4051{
4052#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4053 if (sqe->ioprio || sqe->buf_index || sqe->off)
4054 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004055 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4056 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004057
4058 req->madvise.addr = READ_ONCE(sqe->addr);
4059 req->madvise.len = READ_ONCE(sqe->len);
4060 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4061 return 0;
4062#else
4063 return -EOPNOTSUPP;
4064#endif
4065}
4066
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004067static int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboec1ca7572019-12-25 22:18:28 -07004068{
4069#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4070 struct io_madvise *ma = &req->madvise;
4071 int ret;
4072
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004073 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboec1ca7572019-12-25 22:18:28 -07004074 return -EAGAIN;
4075
Minchan Kim0726b012020-10-17 16:14:50 -07004076 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004077 if (ret < 0)
4078 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004079 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004080 return 0;
4081#else
4082 return -EOPNOTSUPP;
4083#endif
4084}
4085
Jens Axboe4840e412019-12-25 22:03:45 -07004086static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4087{
4088 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4089 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004090 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4091 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004092
4093 req->fadvise.offset = READ_ONCE(sqe->off);
4094 req->fadvise.len = READ_ONCE(sqe->len);
4095 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4096 return 0;
4097}
4098
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004099static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe4840e412019-12-25 22:03:45 -07004100{
4101 struct io_fadvise *fa = &req->fadvise;
4102 int ret;
4103
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004104 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3e694262020-02-01 09:22:49 -07004105 switch (fa->advice) {
4106 case POSIX_FADV_NORMAL:
4107 case POSIX_FADV_RANDOM:
4108 case POSIX_FADV_SEQUENTIAL:
4109 break;
4110 default:
4111 return -EAGAIN;
4112 }
4113 }
Jens Axboe4840e412019-12-25 22:03:45 -07004114
4115 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4116 if (ret < 0)
4117 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004118 io_req_complete(req, ret);
Jens Axboe4840e412019-12-25 22:03:45 -07004119 return 0;
4120}
4121
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004122static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4123{
Jens Axboe6ca56f82020-09-18 16:51:19 -06004124 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004125 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004126 if (sqe->ioprio || sqe->buf_index)
4127 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004128 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004129 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004130
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004131 req->statx.dfd = READ_ONCE(sqe->fd);
4132 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004133 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004134 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4135 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004136
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004137 return 0;
4138}
4139
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004140static int io_statx(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004141{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004142 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004143 int ret;
4144
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004145 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004146 /* only need file table for an actual valid fd */
4147 if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD)
4148 req->flags |= REQ_F_NO_FILE_TABLE;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004149 return -EAGAIN;
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004150 }
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004151
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004152 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4153 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004154
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004155 if (ret < 0)
4156 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004157 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004158 return 0;
4159}
4160
Jens Axboeb5dba592019-12-11 14:02:38 -07004161static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4162{
Jens Axboe14587a462020-09-05 11:36:08 -06004163 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004164 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004165 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4166 sqe->rw_flags || sqe->buf_index)
4167 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004168 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004169 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004170
4171 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboeb5dba592019-12-11 14:02:38 -07004172 return 0;
4173}
4174
Pavel Begunkov889fca72021-02-10 00:03:09 +00004175static int io_close(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb5dba592019-12-11 14:02:38 -07004176{
Jens Axboe9eac1902021-01-19 15:50:37 -07004177 struct files_struct *files = current->files;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004178 struct io_close *close = &req->close;
Jens Axboe9eac1902021-01-19 15:50:37 -07004179 struct fdtable *fdt;
4180 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -07004181 int ret;
4182
Jens Axboe9eac1902021-01-19 15:50:37 -07004183 file = NULL;
4184 ret = -EBADF;
4185 spin_lock(&files->file_lock);
4186 fdt = files_fdtable(files);
4187 if (close->fd >= fdt->max_fds) {
4188 spin_unlock(&files->file_lock);
4189 goto err;
4190 }
4191 file = fdt->fd[close->fd];
4192 if (!file) {
4193 spin_unlock(&files->file_lock);
4194 goto err;
4195 }
4196
4197 if (file->f_op == &io_uring_fops) {
4198 spin_unlock(&files->file_lock);
4199 file = NULL;
4200 goto err;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004201 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004202
4203 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004204 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004205 spin_unlock(&files->file_lock);
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004206 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004207 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004208
Jens Axboe9eac1902021-01-19 15:50:37 -07004209 ret = __close_fd_get_file(close->fd, &file);
4210 spin_unlock(&files->file_lock);
4211 if (ret < 0) {
4212 if (ret == -ENOENT)
4213 ret = -EBADF;
4214 goto err;
4215 }
4216
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004217 /* No ->flush() or already async, safely close from here */
Jens Axboe9eac1902021-01-19 15:50:37 -07004218 ret = filp_close(file, current->files);
4219err:
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004220 if (ret < 0)
4221 req_set_fail_links(req);
Jens Axboe9eac1902021-01-19 15:50:37 -07004222 if (file)
4223 fput(file);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004224 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe1a417f42020-01-31 17:16:48 -07004225 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004226}
4227
Pavel Begunkov1155c762021-02-18 18:29:38 +00004228static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004229{
4230 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004231
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004232 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4233 return -EINVAL;
4234 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4235 return -EINVAL;
4236
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004237 req->sync.off = READ_ONCE(sqe->off);
4238 req->sync.len = READ_ONCE(sqe->len);
4239 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004240 return 0;
4241}
4242
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004243static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004244{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004245 int ret;
4246
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004247 /* sync_file_range always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004248 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004249 return -EAGAIN;
4250
Jens Axboe9adbd452019-12-20 08:45:55 -07004251 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004252 req->sync.flags);
4253 if (ret < 0)
4254 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004255 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004256 return 0;
4257}
4258
YueHaibing469956e2020-03-04 15:53:52 +08004259#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004260static int io_setup_async_msg(struct io_kiocb *req,
4261 struct io_async_msghdr *kmsg)
4262{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004263 struct io_async_msghdr *async_msg = req->async_data;
4264
4265 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004266 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004267 if (io_alloc_async_data(req)) {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004268 kfree(kmsg->free_iov);
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004269 return -ENOMEM;
4270 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004271 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004272 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004273 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov2a780802021-02-05 00:57:58 +00004274 async_msg->msg.msg_name = &async_msg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004275 /* if were using fast_iov, set it to the new one */
4276 if (!async_msg->free_iov)
4277 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
4278
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004279 return -EAGAIN;
4280}
4281
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004282static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4283 struct io_async_msghdr *iomsg)
4284{
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004285 iomsg->msg.msg_name = &iomsg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004286 iomsg->free_iov = iomsg->fast_iov;
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004287 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004288 req->sr_msg.msg_flags, &iomsg->free_iov);
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004289}
4290
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004291static int io_sendmsg_prep_async(struct io_kiocb *req)
4292{
4293 int ret;
4294
4295 if (!io_op_defs[req->opcode].needs_async_data)
4296 return 0;
4297 ret = io_sendmsg_copy_hdr(req, req->async_data);
4298 if (!ret)
4299 req->flags |= REQ_F_NEED_CLEANUP;
4300 return ret;
4301}
4302
Jens Axboe3529d8c2019-12-19 18:24:38 -07004303static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004304{
Jens Axboee47293f2019-12-20 08:58:21 -07004305 struct io_sr_msg *sr = &req->sr_msg;
Jens Axboe03b12302019-12-02 18:50:25 -07004306
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004307 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4308 return -EINVAL;
4309
Jens Axboee47293f2019-12-20 08:58:21 -07004310 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004311 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004312 sr->len = READ_ONCE(sqe->len);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004313
Jens Axboed8768362020-02-27 14:17:49 -07004314#ifdef CONFIG_COMPAT
4315 if (req->ctx->compat)
4316 sr->msg_flags |= MSG_CMSG_COMPAT;
4317#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004318 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004319}
4320
Pavel Begunkov889fca72021-02-10 00:03:09 +00004321static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004322{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004323 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004324 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004325 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004326 int ret;
4327
Florent Revestdba4a922020-12-04 12:36:04 +01004328 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004329 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004330 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004331
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004332 kmsg = req->async_data;
4333 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004334 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004335 if (ret)
4336 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004337 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004338 }
4339
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004340 flags = req->sr_msg.msg_flags;
4341 if (flags & MSG_DONTWAIT)
4342 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004343 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004344 flags |= MSG_DONTWAIT;
4345
4346 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004347 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004348 return io_setup_async_msg(req, kmsg);
4349 if (ret == -ERESTARTSYS)
4350 ret = -EINTR;
4351
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004352 /* fast path, check for non-NULL to avoid function call */
4353 if (kmsg->free_iov)
4354 kfree(kmsg->free_iov);
Jens Axboe03b12302019-12-02 18:50:25 -07004355 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboefddafac2020-01-04 20:19:44 -07004356 if (ret < 0)
4357 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004358 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboefddafac2020-01-04 20:19:44 -07004359 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004360}
4361
Pavel Begunkov889fca72021-02-10 00:03:09 +00004362static int io_send(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004363{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004364 struct io_sr_msg *sr = &req->sr_msg;
4365 struct msghdr msg;
4366 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004367 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004368 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004369 int ret;
4370
Florent Revestdba4a922020-12-04 12:36:04 +01004371 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004372 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004373 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004374
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004375 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4376 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004377 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004378
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004379 msg.msg_name = NULL;
4380 msg.msg_control = NULL;
4381 msg.msg_controllen = 0;
4382 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004383
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004384 flags = req->sr_msg.msg_flags;
4385 if (flags & MSG_DONTWAIT)
4386 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004387 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004388 flags |= MSG_DONTWAIT;
Jens Axboe03b12302019-12-02 18:50:25 -07004389
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004390 msg.msg_flags = flags;
4391 ret = sock_sendmsg(sock, &msg);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004392 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004393 return -EAGAIN;
4394 if (ret == -ERESTARTSYS)
4395 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004396
Jens Axboe03b12302019-12-02 18:50:25 -07004397 if (ret < 0)
4398 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004399 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe03b12302019-12-02 18:50:25 -07004400 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004401}
4402
Pavel Begunkov1400e692020-07-12 20:41:05 +03004403static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4404 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004405{
4406 struct io_sr_msg *sr = &req->sr_msg;
4407 struct iovec __user *uiov;
4408 size_t iov_len;
4409 int ret;
4410
Pavel Begunkov1400e692020-07-12 20:41:05 +03004411 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4412 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004413 if (ret)
4414 return ret;
4415
4416 if (req->flags & REQ_F_BUFFER_SELECT) {
4417 if (iov_len > 1)
4418 return -EINVAL;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004419 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004420 return -EFAULT;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004421 sr->len = iomsg->fast_iov[0].iov_len;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004422 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004423 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004424 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004425 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004426 &iomsg->free_iov, &iomsg->msg.msg_iter,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004427 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004428 if (ret > 0)
4429 ret = 0;
4430 }
4431
4432 return ret;
4433}
4434
4435#ifdef CONFIG_COMPAT
4436static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004437 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004438{
4439 struct compat_msghdr __user *msg_compat;
4440 struct io_sr_msg *sr = &req->sr_msg;
4441 struct compat_iovec __user *uiov;
4442 compat_uptr_t ptr;
4443 compat_size_t len;
4444 int ret;
4445
Pavel Begunkov270a5942020-07-12 20:41:04 +03004446 msg_compat = (struct compat_msghdr __user *) sr->umsg;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004447 ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr,
Jens Axboe52de1fe2020-02-27 10:15:42 -07004448 &ptr, &len);
4449 if (ret)
4450 return ret;
4451
4452 uiov = compat_ptr(ptr);
4453 if (req->flags & REQ_F_BUFFER_SELECT) {
4454 compat_ssize_t clen;
4455
4456 if (len > 1)
4457 return -EINVAL;
4458 if (!access_ok(uiov, sizeof(*uiov)))
4459 return -EFAULT;
4460 if (__get_user(clen, &uiov->iov_len))
4461 return -EFAULT;
4462 if (clen < 0)
4463 return -EINVAL;
Pavel Begunkov2d280bc2020-11-29 18:33:32 +00004464 sr->len = clen;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004465 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004466 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004467 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004468 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004469 UIO_FASTIOV, &iomsg->free_iov,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004470 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004471 if (ret < 0)
4472 return ret;
4473 }
4474
4475 return 0;
4476}
Jens Axboe03b12302019-12-02 18:50:25 -07004477#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004478
Pavel Begunkov1400e692020-07-12 20:41:05 +03004479static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4480 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004481{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004482 iomsg->msg.msg_name = &iomsg->addr;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004483
4484#ifdef CONFIG_COMPAT
4485 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004486 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004487#endif
4488
Pavel Begunkov1400e692020-07-12 20:41:05 +03004489 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004490}
4491
Jens Axboebcda7ba2020-02-23 16:42:51 -07004492static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004493 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004494{
4495 struct io_sr_msg *sr = &req->sr_msg;
4496 struct io_buffer *kbuf;
4497
Jens Axboebcda7ba2020-02-23 16:42:51 -07004498 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4499 if (IS_ERR(kbuf))
4500 return kbuf;
4501
4502 sr->kbuf = kbuf;
4503 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004504 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004505}
4506
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004507static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4508{
4509 return io_put_kbuf(req, req->sr_msg.kbuf);
4510}
4511
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004512static int io_recvmsg_prep_async(struct io_kiocb *req)
Jens Axboe03b12302019-12-02 18:50:25 -07004513{
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004514 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004515
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004516 if (!io_op_defs[req->opcode].needs_async_data)
4517 return 0;
4518 ret = io_recvmsg_copy_hdr(req, req->async_data);
4519 if (!ret)
4520 req->flags |= REQ_F_NEED_CLEANUP;
4521 return ret;
4522}
4523
4524static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4525{
4526 struct io_sr_msg *sr = &req->sr_msg;
4527
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004528 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4529 return -EINVAL;
4530
Jens Axboe3529d8c2019-12-19 18:24:38 -07004531 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004532 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004533 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004534 sr->bgid = READ_ONCE(sqe->buf_group);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004535
Jens Axboed8768362020-02-27 14:17:49 -07004536#ifdef CONFIG_COMPAT
4537 if (req->ctx->compat)
4538 sr->msg_flags |= MSG_CMSG_COMPAT;
4539#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004540 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004541}
4542
Pavel Begunkov889fca72021-02-10 00:03:09 +00004543static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004544{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004545 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004546 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004547 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004548 unsigned flags;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004549 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004550 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004551
Florent Revestdba4a922020-12-04 12:36:04 +01004552 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004553 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004554 return -ENOTSOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004555
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004556 kmsg = req->async_data;
4557 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004558 ret = io_recvmsg_copy_hdr(req, &iomsg);
4559 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004560 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004561 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004562 }
4563
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004564 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004565 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004566 if (IS_ERR(kbuf))
4567 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004568 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004569 kmsg->fast_iov[0].iov_len = req->sr_msg.len;
4570 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov,
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004571 1, req->sr_msg.len);
4572 }
4573
4574 flags = req->sr_msg.msg_flags;
4575 if (flags & MSG_DONTWAIT)
4576 req->flags |= REQ_F_NOWAIT;
4577 else if (force_nonblock)
4578 flags |= MSG_DONTWAIT;
4579
4580 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4581 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004582 if (force_nonblock && ret == -EAGAIN)
4583 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004584 if (ret == -ERESTARTSYS)
4585 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004586
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004587 if (req->flags & REQ_F_BUFFER_SELECTED)
4588 cflags = io_put_recv_kbuf(req);
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004589 /* fast path, check for non-NULL to avoid function call */
4590 if (kmsg->free_iov)
4591 kfree(kmsg->free_iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004592 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004593 if (ret < 0)
4594 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004595 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004596 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004597}
4598
Pavel Begunkov889fca72021-02-10 00:03:09 +00004599static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefddafac2020-01-04 20:19:44 -07004600{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004601 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004602 struct io_sr_msg *sr = &req->sr_msg;
4603 struct msghdr msg;
4604 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004605 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004606 struct iovec iov;
4607 unsigned flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004608 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004609 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004610
Florent Revestdba4a922020-12-04 12:36:04 +01004611 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004612 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004613 return -ENOTSOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004614
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004615 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004616 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004617 if (IS_ERR(kbuf))
4618 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004619 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004620 }
4621
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004622 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004623 if (unlikely(ret))
4624 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004625
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004626 msg.msg_name = NULL;
4627 msg.msg_control = NULL;
4628 msg.msg_controllen = 0;
4629 msg.msg_namelen = 0;
4630 msg.msg_iocb = NULL;
4631 msg.msg_flags = 0;
4632
4633 flags = req->sr_msg.msg_flags;
4634 if (flags & MSG_DONTWAIT)
4635 req->flags |= REQ_F_NOWAIT;
4636 else if (force_nonblock)
4637 flags |= MSG_DONTWAIT;
4638
4639 ret = sock_recvmsg(sock, &msg, flags);
4640 if (force_nonblock && ret == -EAGAIN)
4641 return -EAGAIN;
4642 if (ret == -ERESTARTSYS)
4643 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004644out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004645 if (req->flags & REQ_F_BUFFER_SELECTED)
4646 cflags = io_put_recv_kbuf(req);
Jens Axboefddafac2020-01-04 20:19:44 -07004647 if (ret < 0)
4648 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004649 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboefddafac2020-01-04 20:19:44 -07004650 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004651}
4652
Jens Axboe3529d8c2019-12-19 18:24:38 -07004653static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004654{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004655 struct io_accept *accept = &req->accept;
4656
Jens Axboe14587a462020-09-05 11:36:08 -06004657 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe17f2fe32019-10-17 14:42:58 -06004658 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05004659 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004660 return -EINVAL;
4661
Jens Axboed55e5f52019-12-11 16:12:15 -07004662 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4663 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004664 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06004665 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004666 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004667}
Jens Axboe17f2fe32019-10-17 14:42:58 -06004668
Pavel Begunkov889fca72021-02-10 00:03:09 +00004669static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004670{
4671 struct io_accept *accept = &req->accept;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004672 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004673 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004674 int ret;
4675
Jiufei Xuee697dee2020-06-10 13:41:59 +08004676 if (req->file->f_flags & O_NONBLOCK)
4677 req->flags |= REQ_F_NOWAIT;
4678
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004679 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06004680 accept->addr_len, accept->flags,
4681 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004682 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004683 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004684 if (ret < 0) {
4685 if (ret == -ERESTARTSYS)
4686 ret = -EINTR;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004687 req_set_fail_links(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004688 }
Pavel Begunkov889fca72021-02-10 00:03:09 +00004689 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe17f2fe32019-10-17 14:42:58 -06004690 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004691}
4692
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004693static int io_connect_prep_async(struct io_kiocb *req)
4694{
4695 struct io_async_connect *io = req->async_data;
4696 struct io_connect *conn = &req->connect;
4697
4698 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
4699}
4700
Jens Axboe3529d8c2019-12-19 18:24:38 -07004701static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07004702{
Jens Axboe3529d8c2019-12-19 18:24:38 -07004703 struct io_connect *conn = &req->connect;
Jens Axboef499a022019-12-02 16:28:46 -07004704
Jens Axboe14587a462020-09-05 11:36:08 -06004705 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004706 return -EINVAL;
4707 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
4708 return -EINVAL;
4709
Jens Axboe3529d8c2019-12-19 18:24:38 -07004710 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4711 conn->addr_len = READ_ONCE(sqe->addr2);
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004712 return 0;
Jens Axboef499a022019-12-02 16:28:46 -07004713}
4714
Pavel Begunkov889fca72021-02-10 00:03:09 +00004715static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboef8e85cf2019-11-23 14:24:24 -07004716{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004717 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004718 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004719 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004720 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004721
Jens Axboee8c2bc12020-08-15 18:44:09 -07004722 if (req->async_data) {
4723 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004724 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07004725 ret = move_addr_to_kernel(req->connect.addr,
4726 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004727 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07004728 if (ret)
4729 goto out;
4730 io = &__io;
4731 }
4732
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004733 file_flags = force_nonblock ? O_NONBLOCK : 0;
4734
Jens Axboee8c2bc12020-08-15 18:44:09 -07004735 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004736 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07004737 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07004738 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07004739 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004740 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07004741 ret = -ENOMEM;
4742 goto out;
4743 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004744 io = req->async_data;
4745 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07004746 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07004747 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07004748 if (ret == -ERESTARTSYS)
4749 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07004750out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004751 if (ret < 0)
4752 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004753 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboef8e85cf2019-11-23 14:24:24 -07004754 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004755}
YueHaibing469956e2020-03-04 15:53:52 +08004756#else /* !CONFIG_NET */
Jens Axboe99a10082021-02-19 09:35:19 -07004757#define IO_NETOP_FN(op) \
4758static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \
4759{ \
4760 return -EOPNOTSUPP; \
Jens Axboef8e85cf2019-11-23 14:24:24 -07004761}
4762
Jens Axboe99a10082021-02-19 09:35:19 -07004763#define IO_NETOP_PREP(op) \
4764IO_NETOP_FN(op) \
4765static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \
4766{ \
4767 return -EOPNOTSUPP; \
4768} \
4769
4770#define IO_NETOP_PREP_ASYNC(op) \
4771IO_NETOP_PREP(op) \
4772static int io_##op##_prep_async(struct io_kiocb *req) \
4773{ \
4774 return -EOPNOTSUPP; \
YueHaibing469956e2020-03-04 15:53:52 +08004775}
4776
Jens Axboe99a10082021-02-19 09:35:19 -07004777IO_NETOP_PREP_ASYNC(sendmsg);
4778IO_NETOP_PREP_ASYNC(recvmsg);
4779IO_NETOP_PREP_ASYNC(connect);
4780IO_NETOP_PREP(accept);
4781IO_NETOP_FN(send);
4782IO_NETOP_FN(recv);
YueHaibing469956e2020-03-04 15:53:52 +08004783#endif /* CONFIG_NET */
Jens Axboe17f2fe32019-10-17 14:42:58 -06004784
Jens Axboed7718a92020-02-14 22:23:12 -07004785struct io_poll_table {
4786 struct poll_table_struct pt;
4787 struct io_kiocb *req;
4788 int error;
4789};
4790
Jens Axboed7718a92020-02-14 22:23:12 -07004791static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
4792 __poll_t mask, task_work_func_t func)
4793{
Jens Axboeaa96bf82020-04-03 11:26:26 -06004794 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07004795
4796 /* for instances that support it check for an event match first: */
4797 if (mask && !(mask & poll->events))
4798 return 0;
4799
4800 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
4801
4802 list_del_init(&poll->wait.entry);
4803
Jens Axboed7718a92020-02-14 22:23:12 -07004804 req->result = mask;
Jens Axboe7cbf1722021-02-10 00:03:20 +00004805 req->task_work.func = func;
Jens Axboe6d816e02020-08-11 08:04:14 -06004806 percpu_ref_get(&req->ctx->refs);
4807
Jens Axboed7718a92020-02-14 22:23:12 -07004808 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06004809 * If this fails, then the task is exiting. When a task exits, the
4810 * work gets canceled, so just cancel this request as well instead
4811 * of executing it. We can't safely execute it anyway, as we may not
4812 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07004813 */
Jens Axboe355fb9e2020-10-22 20:19:35 -06004814 ret = io_req_task_work_add(req);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004815 if (unlikely(ret)) {
Jens Axboee3aabf92020-05-18 11:04:17 -06004816 WRITE_ONCE(poll->canceled, true);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00004817 io_req_task_work_add_fallback(req, func);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004818 }
Jens Axboed7718a92020-02-14 22:23:12 -07004819 return 1;
4820}
4821
Jens Axboe74ce6ce2020-04-13 11:09:12 -06004822static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
4823 __acquires(&req->ctx->completion_lock)
4824{
4825 struct io_ring_ctx *ctx = req->ctx;
4826
4827 if (!req->result && !READ_ONCE(poll->canceled)) {
4828 struct poll_table_struct pt = { ._key = poll->events };
4829
4830 req->result = vfs_poll(req->file, &pt) & poll->events;
4831 }
4832
4833 spin_lock_irq(&ctx->completion_lock);
4834 if (!req->result && !READ_ONCE(poll->canceled)) {
4835 add_wait_queue(poll->head, &poll->wait);
4836 return true;
4837 }
4838
4839 return false;
4840}
4841
Jens Axboed4e7cd32020-08-15 11:44:50 -07004842static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06004843{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004844 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07004845 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07004846 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004847 return req->apoll->double_poll;
4848}
4849
4850static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
4851{
4852 if (req->opcode == IORING_OP_POLL_ADD)
4853 return &req->poll;
4854 return &req->apoll->poll;
4855}
4856
4857static void io_poll_remove_double(struct io_kiocb *req)
4858{
4859 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004860
4861 lockdep_assert_held(&req->ctx->completion_lock);
4862
4863 if (poll && poll->head) {
4864 struct wait_queue_head *head = poll->head;
4865
4866 spin_lock(&head->lock);
4867 list_del_init(&poll->wait.entry);
4868 if (poll->wait.private)
4869 refcount_dec(&req->refs);
4870 poll->head = NULL;
4871 spin_unlock(&head->lock);
4872 }
4873}
4874
4875static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error)
4876{
4877 struct io_ring_ctx *ctx = req->ctx;
4878
Jens Axboed4e7cd32020-08-15 11:44:50 -07004879 io_poll_remove_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004880 req->poll.done = true;
4881 io_cqring_fill_event(req, error ? error : mangle_poll(mask));
4882 io_commit_cqring(ctx);
4883}
4884
Jens Axboe18bceab2020-05-15 11:56:54 -06004885static void io_poll_task_func(struct callback_head *cb)
4886{
4887 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06004888 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004889 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06004890
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004891 if (io_poll_rewait(req, &req->poll)) {
4892 spin_unlock_irq(&ctx->completion_lock);
4893 } else {
4894 hash_del(&req->hash_node);
4895 io_poll_complete(req, req->result, 0);
4896 spin_unlock_irq(&ctx->completion_lock);
4897
4898 nxt = io_put_req_find_next(req);
4899 io_cqring_ev_posted(ctx);
4900 if (nxt)
4901 __io_req_task_submit(nxt);
4902 }
4903
Jens Axboe6d816e02020-08-11 08:04:14 -06004904 percpu_ref_put(&ctx->refs);
Jens Axboe18bceab2020-05-15 11:56:54 -06004905}
4906
4907static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
4908 int sync, void *key)
4909{
4910 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004911 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004912 __poll_t mask = key_to_poll(key);
4913
4914 /* for instances that support it check for an event match first: */
4915 if (mask && !(mask & poll->events))
4916 return 0;
4917
Jens Axboe8706e042020-09-28 08:38:54 -06004918 list_del_init(&wait->entry);
4919
Jens Axboe807abcb2020-07-17 17:09:27 -06004920 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004921 bool done;
4922
Jens Axboe807abcb2020-07-17 17:09:27 -06004923 spin_lock(&poll->head->lock);
4924 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06004925 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06004926 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07004927 /* make sure double remove sees this as being gone */
4928 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06004929 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06004930 if (!done) {
4931 /* use wait func handler, so it matches the rq type */
4932 poll->wait.func(&poll->wait, mode, sync, key);
4933 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004934 }
4935 refcount_dec(&req->refs);
4936 return 1;
4937}
4938
4939static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
4940 wait_queue_func_t wake_func)
4941{
4942 poll->head = NULL;
4943 poll->done = false;
4944 poll->canceled = false;
4945 poll->events = events;
4946 INIT_LIST_HEAD(&poll->wait.entry);
4947 init_waitqueue_func_entry(&poll->wait, wake_func);
4948}
4949
4950static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06004951 struct wait_queue_head *head,
4952 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06004953{
4954 struct io_kiocb *req = pt->req;
4955
4956 /*
4957 * If poll->head is already set, it's because the file being polled
4958 * uses multiple waitqueues for poll handling (eg one for read, one
4959 * for write). Setup a separate io_poll_iocb if this happens.
4960 */
4961 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01004962 struct io_poll_iocb *poll_one = poll;
4963
Jens Axboe18bceab2020-05-15 11:56:54 -06004964 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06004965 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004966 pt->error = -EINVAL;
4967 return;
4968 }
Jens Axboe1c3b3e62021-02-28 16:07:30 -07004969 /* double add on the same waitqueue head, ignore */
4970 if (poll->head == head)
4971 return;
Jens Axboe18bceab2020-05-15 11:56:54 -06004972 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
4973 if (!poll) {
4974 pt->error = -ENOMEM;
4975 return;
4976 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01004977 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboe18bceab2020-05-15 11:56:54 -06004978 refcount_inc(&req->refs);
4979 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06004980 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06004981 }
4982
4983 pt->error = 0;
4984 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08004985
4986 if (poll->events & EPOLLEXCLUSIVE)
4987 add_wait_queue_exclusive(head, &poll->wait);
4988 else
4989 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06004990}
4991
4992static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
4993 struct poll_table_struct *p)
4994{
4995 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06004996 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06004997
Jens Axboe807abcb2020-07-17 17:09:27 -06004998 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06004999}
5000
Jens Axboed7718a92020-02-14 22:23:12 -07005001static void io_async_task_func(struct callback_head *cb)
5002{
5003 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5004 struct async_poll *apoll = req->apoll;
5005 struct io_ring_ctx *ctx = req->ctx;
5006
5007 trace_io_uring_task_run(req->ctx, req->opcode, req->user_data);
5008
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005009 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005010 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe6d816e02020-08-11 08:04:14 -06005011 percpu_ref_put(&ctx->refs);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005012 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005013 }
5014
Jens Axboe31067252020-05-17 17:43:31 -06005015 /* If req is still hashed, it cannot have been canceled. Don't check. */
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005016 if (hash_hashed(&req->hash_node))
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005017 hash_del(&req->hash_node);
Jens Axboe2bae0472020-04-13 11:16:34 -06005018
Jens Axboed4e7cd32020-08-15 11:44:50 -07005019 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005020 spin_unlock_irq(&ctx->completion_lock);
5021
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005022 if (!READ_ONCE(apoll->poll.canceled))
5023 __io_req_task_submit(req);
5024 else
5025 __io_req_task_cancel(req, -ECANCELED);
Dan Carpenteraa340842020-07-08 21:47:11 +03005026
Jens Axboe6d816e02020-08-11 08:04:14 -06005027 percpu_ref_put(&ctx->refs);
Jens Axboe807abcb2020-07-17 17:09:27 -06005028 kfree(apoll->double_poll);
Jens Axboe31067252020-05-17 17:43:31 -06005029 kfree(apoll);
Jens Axboed7718a92020-02-14 22:23:12 -07005030}
5031
5032static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5033 void *key)
5034{
5035 struct io_kiocb *req = wait->private;
5036 struct io_poll_iocb *poll = &req->apoll->poll;
5037
5038 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5039 key_to_poll(key));
5040
5041 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5042}
5043
5044static void io_poll_req_insert(struct io_kiocb *req)
5045{
5046 struct io_ring_ctx *ctx = req->ctx;
5047 struct hlist_head *list;
5048
5049 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5050 hlist_add_head(&req->hash_node, list);
5051}
5052
5053static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5054 struct io_poll_iocb *poll,
5055 struct io_poll_table *ipt, __poll_t mask,
5056 wait_queue_func_t wake_func)
5057 __acquires(&ctx->completion_lock)
5058{
5059 struct io_ring_ctx *ctx = req->ctx;
5060 bool cancel = false;
5061
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005062 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005063 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005064 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005065 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005066
5067 ipt->pt._key = mask;
5068 ipt->req = req;
5069 ipt->error = -EINVAL;
5070
Jens Axboed7718a92020-02-14 22:23:12 -07005071 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5072
5073 spin_lock_irq(&ctx->completion_lock);
5074 if (likely(poll->head)) {
5075 spin_lock(&poll->head->lock);
5076 if (unlikely(list_empty(&poll->wait.entry))) {
5077 if (ipt->error)
5078 cancel = true;
5079 ipt->error = 0;
5080 mask = 0;
5081 }
5082 if (mask || ipt->error)
5083 list_del_init(&poll->wait.entry);
5084 else if (cancel)
5085 WRITE_ONCE(poll->canceled, true);
5086 else if (!poll->done) /* actually waiting for an event */
5087 io_poll_req_insert(req);
5088 spin_unlock(&poll->head->lock);
5089 }
5090
5091 return mask;
5092}
5093
5094static bool io_arm_poll_handler(struct io_kiocb *req)
5095{
5096 const struct io_op_def *def = &io_op_defs[req->opcode];
5097 struct io_ring_ctx *ctx = req->ctx;
5098 struct async_poll *apoll;
5099 struct io_poll_table ipt;
5100 __poll_t mask, ret;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005101 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005102
5103 if (!req->file || !file_can_poll(req->file))
5104 return false;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005105 if (req->flags & REQ_F_POLLED)
Jens Axboed7718a92020-02-14 22:23:12 -07005106 return false;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005107 if (def->pollin)
5108 rw = READ;
5109 else if (def->pollout)
5110 rw = WRITE;
5111 else
5112 return false;
5113 /* if we can't nonblock try, then no point in arming a poll handler */
5114 if (!io_file_supports_async(req->file, rw))
Jens Axboed7718a92020-02-14 22:23:12 -07005115 return false;
5116
5117 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5118 if (unlikely(!apoll))
5119 return false;
Jens Axboe807abcb2020-07-17 17:09:27 -06005120 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005121
5122 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005123 req->apoll = apoll;
Jens Axboed7718a92020-02-14 22:23:12 -07005124
Nathan Chancellor8755d972020-03-02 16:01:19 -07005125 mask = 0;
Jens Axboed7718a92020-02-14 22:23:12 -07005126 if (def->pollin)
Nathan Chancellor8755d972020-03-02 16:01:19 -07005127 mask |= POLLIN | POLLRDNORM;
Jens Axboed7718a92020-02-14 22:23:12 -07005128 if (def->pollout)
5129 mask |= POLLOUT | POLLWRNORM;
Luke Hsiao901341b2020-08-21 21:41:05 -07005130
5131 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5132 if ((req->opcode == IORING_OP_RECVMSG) &&
5133 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5134 mask &= ~POLLIN;
5135
Jens Axboed7718a92020-02-14 22:23:12 -07005136 mask |= POLLERR | POLLPRI;
5137
5138 ipt.pt._qproc = io_async_queue_proc;
5139
5140 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5141 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005142 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005143 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005144 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe807abcb2020-07-17 17:09:27 -06005145 kfree(apoll->double_poll);
Jens Axboed7718a92020-02-14 22:23:12 -07005146 kfree(apoll);
5147 return false;
5148 }
5149 spin_unlock_irq(&ctx->completion_lock);
5150 trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
5151 apoll->poll.events);
5152 return true;
5153}
5154
5155static bool __io_poll_remove_one(struct io_kiocb *req,
5156 struct io_poll_iocb *poll)
5157{
Jens Axboeb41e9852020-02-17 09:52:41 -07005158 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005159
5160 spin_lock(&poll->head->lock);
5161 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005162 if (!list_empty(&poll->wait.entry)) {
5163 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005164 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005165 }
5166 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005167 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005168 return do_complete;
5169}
5170
5171static bool io_poll_remove_one(struct io_kiocb *req)
5172{
5173 bool do_complete;
5174
Jens Axboed4e7cd32020-08-15 11:44:50 -07005175 io_poll_remove_double(req);
5176
Jens Axboed7718a92020-02-14 22:23:12 -07005177 if (req->opcode == IORING_OP_POLL_ADD) {
5178 do_complete = __io_poll_remove_one(req, &req->poll);
5179 } else {
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005180 struct async_poll *apoll = req->apoll;
5181
Jens Axboed7718a92020-02-14 22:23:12 -07005182 /* non-poll requests have submit ref still */
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005183 do_complete = __io_poll_remove_one(req, &apoll->poll);
5184 if (do_complete) {
Jens Axboed7718a92020-02-14 22:23:12 -07005185 io_put_req(req);
Jens Axboe807abcb2020-07-17 17:09:27 -06005186 kfree(apoll->double_poll);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005187 kfree(apoll);
5188 }
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005189 }
5190
Jens Axboeb41e9852020-02-17 09:52:41 -07005191 if (do_complete) {
5192 io_cqring_fill_event(req, -ECANCELED);
5193 io_commit_cqring(req->ctx);
Jens Axboef254ac02020-08-12 17:33:30 -06005194 req_set_fail_links(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005195 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005196 }
5197
5198 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005199}
5200
Jens Axboe76e1b642020-09-26 15:05:03 -06005201/*
5202 * Returns true if we found and killed one or more poll requests
5203 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00005204static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
5205 struct files_struct *files)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005206{
Jens Axboe78076bb2019-12-04 19:56:40 -07005207 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005208 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005209 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005210
5211 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005212 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5213 struct hlist_head *list;
5214
5215 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005216 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00005217 if (io_match_task(req, tsk, files))
Jens Axboef3606e32020-09-22 08:18:24 -06005218 posted += io_poll_remove_one(req);
5219 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005220 }
5221 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005222
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005223 if (posted)
5224 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005225
5226 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005227}
5228
Jens Axboe47f46762019-11-09 17:43:02 -07005229static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
5230{
Jens Axboe78076bb2019-12-04 19:56:40 -07005231 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005232 struct io_kiocb *req;
5233
Jens Axboe78076bb2019-12-04 19:56:40 -07005234 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5235 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005236 if (sqe_addr != req->user_data)
5237 continue;
5238 if (io_poll_remove_one(req))
Jens Axboeeac406c2019-11-14 12:09:58 -07005239 return 0;
Jens Axboeb41e9852020-02-17 09:52:41 -07005240 return -EALREADY;
Jens Axboe47f46762019-11-09 17:43:02 -07005241 }
5242
5243 return -ENOENT;
5244}
5245
Jens Axboe3529d8c2019-12-19 18:24:38 -07005246static int io_poll_remove_prep(struct io_kiocb *req,
5247 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005248{
Jens Axboe221c5eb2019-01-17 09:41:58 -07005249 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5250 return -EINVAL;
5251 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
5252 sqe->poll_events)
5253 return -EINVAL;
5254
Pavel Begunkov018043b2020-10-27 23:17:18 +00005255 req->poll_remove.addr = READ_ONCE(sqe->addr);
Jens Axboe0969e782019-12-17 18:40:57 -07005256 return 0;
5257}
5258
5259/*
5260 * Find a running poll command that matches one specified in sqe->addr,
5261 * and remove it if found.
5262 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005263static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005264{
5265 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe0969e782019-12-17 18:40:57 -07005266 int ret;
5267
Jens Axboe221c5eb2019-01-17 09:41:58 -07005268 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov018043b2020-10-27 23:17:18 +00005269 ret = io_poll_cancel(ctx, req->poll_remove.addr);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005270 spin_unlock_irq(&ctx->completion_lock);
5271
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005272 if (ret < 0)
5273 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06005274 io_req_complete(req, ret);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005275 return 0;
5276}
5277
Jens Axboe221c5eb2019-01-17 09:41:58 -07005278static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5279 void *key)
5280{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005281 struct io_kiocb *req = wait->private;
5282 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005283
Jens Axboed7718a92020-02-14 22:23:12 -07005284 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005285}
5286
Jens Axboe221c5eb2019-01-17 09:41:58 -07005287static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5288 struct poll_table_struct *p)
5289{
5290 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5291
Jens Axboee8c2bc12020-08-15 18:44:09 -07005292 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c2019-11-14 12:09:58 -07005293}
5294
Jens Axboe3529d8c2019-12-19 18:24:38 -07005295static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005296{
5297 struct io_poll_iocb *poll = &req->poll;
Jiufei Xue5769a352020-06-17 17:53:55 +08005298 u32 events;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005299
5300 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5301 return -EINVAL;
5302 if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
5303 return -EINVAL;
5304
Jiufei Xue5769a352020-06-17 17:53:55 +08005305 events = READ_ONCE(sqe->poll32_events);
5306#ifdef __BIG_ENDIAN
5307 events = swahw32(events);
5308#endif
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005309 poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP |
5310 (events & EPOLLEXCLUSIVE);
Jens Axboe0969e782019-12-17 18:40:57 -07005311 return 0;
5312}
5313
Pavel Begunkov61e98202021-02-10 00:03:08 +00005314static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005315{
5316 struct io_poll_iocb *poll = &req->poll;
5317 struct io_ring_ctx *ctx = req->ctx;
5318 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005319 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005320
Jens Axboed7718a92020-02-14 22:23:12 -07005321 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005322
Jens Axboed7718a92020-02-14 22:23:12 -07005323 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5324 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005325
Jens Axboe8c838782019-03-12 15:48:16 -06005326 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005327 ipt.error = 0;
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005328 io_poll_complete(req, mask, 0);
Jens Axboe8c838782019-03-12 15:48:16 -06005329 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005330 spin_unlock_irq(&ctx->completion_lock);
5331
Jens Axboe8c838782019-03-12 15:48:16 -06005332 if (mask) {
5333 io_cqring_ev_posted(ctx);
Pavel Begunkov014db002020-03-03 21:33:12 +03005334 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005335 }
Jens Axboe8c838782019-03-12 15:48:16 -06005336 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005337}
5338
Jens Axboe5262f562019-09-17 12:26:57 -06005339static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5340{
Jens Axboead8a48a2019-11-15 08:49:11 -07005341 struct io_timeout_data *data = container_of(timer,
5342 struct io_timeout_data, timer);
5343 struct io_kiocb *req = data->req;
5344 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005345 unsigned long flags;
5346
Jens Axboe5262f562019-09-17 12:26:57 -06005347 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005348 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005349 atomic_set(&req->ctx->cq_timeouts,
5350 atomic_read(&req->ctx->cq_timeouts) + 1);
5351
Jens Axboe78e19bb2019-11-06 15:21:34 -07005352 io_cqring_fill_event(req, -ETIME);
Jens Axboe5262f562019-09-17 12:26:57 -06005353 io_commit_cqring(ctx);
5354 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5355
5356 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005357 req_set_fail_links(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005358 io_put_req(req);
5359 return HRTIMER_NORESTART;
5360}
5361
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005362static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx,
5363 __u64 user_data)
Jens Axboe47f46762019-11-09 17:43:02 -07005364{
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005365 struct io_timeout_data *io;
Jens Axboef254ac02020-08-12 17:33:30 -06005366 struct io_kiocb *req;
5367 int ret = -ENOENT;
5368
5369 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
5370 if (user_data == req->user_data) {
5371 ret = 0;
5372 break;
5373 }
5374 }
5375
5376 if (ret == -ENOENT)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005377 return ERR_PTR(ret);
Jens Axboef254ac02020-08-12 17:33:30 -06005378
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005379 io = req->async_data;
5380 ret = hrtimer_try_to_cancel(&io->timer);
5381 if (ret == -1)
5382 return ERR_PTR(-EALREADY);
5383 list_del_init(&req->timeout.list);
5384 return req;
5385}
5386
5387static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
5388{
5389 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5390
5391 if (IS_ERR(req))
5392 return PTR_ERR(req);
5393
5394 req_set_fail_links(req);
5395 io_cqring_fill_event(req, -ECANCELED);
5396 io_put_req_deferred(req, 1);
5397 return 0;
Jens Axboef254ac02020-08-12 17:33:30 -06005398}
5399
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005400static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
5401 struct timespec64 *ts, enum hrtimer_mode mode)
5402{
5403 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5404 struct io_timeout_data *data;
5405
5406 if (IS_ERR(req))
5407 return PTR_ERR(req);
5408
5409 req->timeout.off = 0; /* noseq */
5410 data = req->async_data;
5411 list_add_tail(&req->timeout.list, &ctx->timeout_list);
5412 hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
5413 data->timer.function = io_timeout_fn;
5414 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
5415 return 0;
Jens Axboe47f46762019-11-09 17:43:02 -07005416}
5417
Jens Axboe3529d8c2019-12-19 18:24:38 -07005418static int io_timeout_remove_prep(struct io_kiocb *req,
5419 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005420{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005421 struct io_timeout_rem *tr = &req->timeout_rem;
5422
Jens Axboeb29472e2019-12-17 18:50:29 -07005423 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5424 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005425 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5426 return -EINVAL;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005427 if (sqe->ioprio || sqe->buf_index || sqe->len)
Jens Axboeb29472e2019-12-17 18:50:29 -07005428 return -EINVAL;
5429
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005430 tr->addr = READ_ONCE(sqe->addr);
5431 tr->flags = READ_ONCE(sqe->timeout_flags);
5432 if (tr->flags & IORING_TIMEOUT_UPDATE) {
5433 if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS))
5434 return -EINVAL;
5435 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2)))
5436 return -EFAULT;
5437 } else if (tr->flags) {
5438 /* timeout removal doesn't support flags */
5439 return -EINVAL;
5440 }
5441
Jens Axboeb29472e2019-12-17 18:50:29 -07005442 return 0;
5443}
5444
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005445static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags)
5446{
5447 return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS
5448 : HRTIMER_MODE_REL;
5449}
5450
Jens Axboe11365042019-10-16 09:08:32 -06005451/*
5452 * Remove or update an existing timeout command
5453 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005454static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe11365042019-10-16 09:08:32 -06005455{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005456 struct io_timeout_rem *tr = &req->timeout_rem;
Jens Axboe11365042019-10-16 09:08:32 -06005457 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005458 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005459
Jens Axboe11365042019-10-16 09:08:32 -06005460 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005461 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE))
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005462 ret = io_timeout_cancel(ctx, tr->addr);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005463 else
5464 ret = io_timeout_update(ctx, tr->addr, &tr->ts,
5465 io_translate_timeout_mode(tr->flags));
Jens Axboe11365042019-10-16 09:08:32 -06005466
Jens Axboe47f46762019-11-09 17:43:02 -07005467 io_cqring_fill_event(req, ret);
Jens Axboe11365042019-10-16 09:08:32 -06005468 io_commit_cqring(ctx);
5469 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005470 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005471 if (ret < 0)
5472 req_set_fail_links(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005473 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005474 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005475}
5476
Jens Axboe3529d8c2019-12-19 18:24:38 -07005477static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005478 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005479{
Jens Axboead8a48a2019-11-15 08:49:11 -07005480 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005481 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005482 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005483
Jens Axboead8a48a2019-11-15 08:49:11 -07005484 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005485 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005486 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005487 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005488 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005489 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005490 flags = READ_ONCE(sqe->timeout_flags);
5491 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005492 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005493
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005494 req->timeout.off = off;
Jens Axboe26a61672019-12-20 09:02:01 -07005495
Jens Axboee8c2bc12020-08-15 18:44:09 -07005496 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005497 return -ENOMEM;
5498
Jens Axboee8c2bc12020-08-15 18:44:09 -07005499 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005500 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005501
5502 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005503 return -EFAULT;
5504
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005505 data->mode = io_translate_timeout_mode(flags);
Jens Axboead8a48a2019-11-15 08:49:11 -07005506 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
Pavel Begunkovdd59a3d2021-03-04 13:59:25 +00005507 io_req_track_inflight(req);
Jens Axboead8a48a2019-11-15 08:49:11 -07005508 return 0;
5509}
5510
Pavel Begunkov61e98202021-02-10 00:03:08 +00005511static int io_timeout(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboead8a48a2019-11-15 08:49:11 -07005512{
Jens Axboead8a48a2019-11-15 08:49:11 -07005513 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005514 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005515 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005516 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005517
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005518 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005519
Jens Axboe5262f562019-09-17 12:26:57 -06005520 /*
5521 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005522 * timeout event to be satisfied. If it isn't set, then this is
5523 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005524 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005525 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005526 entry = ctx->timeout_list.prev;
5527 goto add;
5528 }
Jens Axboe5262f562019-09-17 12:26:57 -06005529
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005530 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5531 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005532
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05005533 /* Update the last seq here in case io_flush_timeouts() hasn't.
5534 * This is safe because ->completion_lock is held, and submissions
5535 * and completions are never mixed in the same ->completion_lock section.
5536 */
5537 ctx->cq_last_tm_flush = tail;
5538
Jens Axboe5262f562019-09-17 12:26:57 -06005539 /*
5540 * Insertion sort, ensuring the first entry in the list is always
5541 * the one we need first.
5542 */
Jens Axboe5262f562019-09-17 12:26:57 -06005543 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005544 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5545 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005546
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005547 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005548 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005549 /* nxt.seq is behind @tail, otherwise would've been completed */
5550 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005551 break;
5552 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005553add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005554 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005555 data->timer.function = io_timeout_fn;
5556 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005557 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005558 return 0;
5559}
5560
Jens Axboe62755e32019-10-28 21:49:21 -06005561static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005562{
Jens Axboe62755e32019-10-28 21:49:21 -06005563 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Jens Axboede0617e2019-04-06 21:51:27 -06005564
Jens Axboe62755e32019-10-28 21:49:21 -06005565 return req->user_data == (unsigned long) data;
5566}
5567
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005568static int io_async_cancel_one(struct io_uring_task *tctx, void *sqe_addr)
Jens Axboe62755e32019-10-28 21:49:21 -06005569{
Jens Axboe62755e32019-10-28 21:49:21 -06005570 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005571 int ret = 0;
5572
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005573 if (!tctx->io_wq)
5574 return -ENOENT;
5575
5576 cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, sqe_addr, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005577 switch (cancel_ret) {
5578 case IO_WQ_CANCEL_OK:
5579 ret = 0;
5580 break;
5581 case IO_WQ_CANCEL_RUNNING:
5582 ret = -EALREADY;
5583 break;
5584 case IO_WQ_CANCEL_NOTFOUND:
5585 ret = -ENOENT;
5586 break;
5587 }
5588
Jens Axboee977d6d2019-11-05 12:39:45 -07005589 return ret;
5590}
5591
Jens Axboe47f46762019-11-09 17:43:02 -07005592static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5593 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005594 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005595{
5596 unsigned long flags;
5597 int ret;
5598
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005599 ret = io_async_cancel_one(req->task->io_uring,
5600 (void *) (unsigned long) sqe_addr);
Jens Axboe47f46762019-11-09 17:43:02 -07005601 if (ret != -ENOENT) {
5602 spin_lock_irqsave(&ctx->completion_lock, flags);
5603 goto done;
5604 }
5605
5606 spin_lock_irqsave(&ctx->completion_lock, flags);
5607 ret = io_timeout_cancel(ctx, sqe_addr);
5608 if (ret != -ENOENT)
5609 goto done;
5610 ret = io_poll_cancel(ctx, sqe_addr);
5611done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005612 if (!ret)
5613 ret = success_ret;
Jens Axboe47f46762019-11-09 17:43:02 -07005614 io_cqring_fill_event(req, ret);
5615 io_commit_cqring(ctx);
5616 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5617 io_cqring_ev_posted(ctx);
5618
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005619 if (ret < 0)
5620 req_set_fail_links(req);
Pavel Begunkov014db002020-03-03 21:33:12 +03005621 io_put_req(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005622}
5623
Jens Axboe3529d8c2019-12-19 18:24:38 -07005624static int io_async_cancel_prep(struct io_kiocb *req,
5625 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07005626{
Jens Axboefbf23842019-12-17 18:45:56 -07005627 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07005628 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005629 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5630 return -EINVAL;
5631 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07005632 return -EINVAL;
5633
Jens Axboefbf23842019-12-17 18:45:56 -07005634 req->cancel.addr = READ_ONCE(sqe->addr);
5635 return 0;
5636}
5637
Pavel Begunkov61e98202021-02-10 00:03:08 +00005638static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefbf23842019-12-17 18:45:56 -07005639{
5640 struct io_ring_ctx *ctx = req->ctx;
Jens Axboefbf23842019-12-17 18:45:56 -07005641
Pavel Begunkov014db002020-03-03 21:33:12 +03005642 io_async_find_and_cancel(ctx, req, req->cancel.addr, 0);
Jens Axboe62755e32019-10-28 21:49:21 -06005643 return 0;
5644}
5645
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005646static int io_rsrc_update_prep(struct io_kiocb *req,
Jens Axboe05f3fb32019-12-09 11:22:50 -07005647 const struct io_uring_sqe *sqe)
5648{
Jens Axboe6ca56f82020-09-18 16:51:19 -06005649 if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL))
5650 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005651 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5652 return -EINVAL;
5653 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005654 return -EINVAL;
5655
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005656 req->rsrc_update.offset = READ_ONCE(sqe->off);
5657 req->rsrc_update.nr_args = READ_ONCE(sqe->len);
5658 if (!req->rsrc_update.nr_args)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005659 return -EINVAL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005660 req->rsrc_update.arg = READ_ONCE(sqe->addr);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005661 return 0;
5662}
5663
Pavel Begunkov889fca72021-02-10 00:03:09 +00005664static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005665{
5666 struct io_ring_ctx *ctx = req->ctx;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005667 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005668 int ret;
5669
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005670 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005671 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005672
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005673 up.offset = req->rsrc_update.offset;
5674 up.data = req->rsrc_update.arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005675
5676 mutex_lock(&ctx->uring_lock);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005677 ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005678 mutex_unlock(&ctx->uring_lock);
5679
5680 if (ret < 0)
5681 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005682 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005683 return 0;
5684}
5685
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005686static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07005687{
Jens Axboed625c6e2019-12-17 19:53:05 -07005688 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07005689 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005690 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07005691 case IORING_OP_READV:
5692 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005693 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005694 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005695 case IORING_OP_WRITEV:
5696 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005697 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005698 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005699 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005700 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005701 case IORING_OP_POLL_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005702 return io_poll_remove_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005703 case IORING_OP_FSYNC:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005704 return io_fsync_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005705 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005706 return io_sfr_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005707 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005708 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005709 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005710 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005711 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005712 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07005713 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005714 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005715 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005716 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07005717 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005718 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07005719 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005720 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005721 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005722 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005723 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005724 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07005725 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005726 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005727 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005728 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07005729 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005730 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005731 case IORING_OP_FILES_UPDATE:
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005732 return io_rsrc_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005733 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005734 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07005735 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005736 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07005737 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005738 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07005739 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005740 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005741 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005742 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005743 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005744 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005745 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005746 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07005747 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005748 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03005749 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005750 return io_tee_prep(req, sqe);
Jens Axboe36f4fa62020-09-05 11:14:22 -06005751 case IORING_OP_SHUTDOWN:
5752 return io_shutdown_prep(req, sqe);
Jens Axboe80a261f2020-09-28 14:23:58 -06005753 case IORING_OP_RENAMEAT:
5754 return io_renameat_prep(req, sqe);
Jens Axboe14a11432020-09-28 14:27:37 -06005755 case IORING_OP_UNLINKAT:
5756 return io_unlinkat_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005757 }
5758
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005759 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
5760 req->opcode);
5761 return-EINVAL;
5762}
5763
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005764static int io_req_prep_async(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005765{
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005766 switch (req->opcode) {
5767 case IORING_OP_READV:
5768 case IORING_OP_READ_FIXED:
5769 case IORING_OP_READ:
5770 return io_rw_prep_async(req, READ);
5771 case IORING_OP_WRITEV:
5772 case IORING_OP_WRITE_FIXED:
5773 case IORING_OP_WRITE:
5774 return io_rw_prep_async(req, WRITE);
5775 case IORING_OP_SENDMSG:
5776 case IORING_OP_SEND:
5777 return io_sendmsg_prep_async(req);
5778 case IORING_OP_RECVMSG:
5779 case IORING_OP_RECV:
5780 return io_recvmsg_prep_async(req);
5781 case IORING_OP_CONNECT:
5782 return io_connect_prep_async(req);
5783 }
5784 return 0;
5785}
5786
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005787static int io_req_defer_prep(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005788{
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005789 if (!io_op_defs[req->opcode].needs_async_data)
Jens Axboe2b188cc2019-01-07 10:46:33 -07005790 return 0;
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005791 /* some opcodes init it during the inital prep */
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005792 if (req->async_data)
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005793 return 0;
5794 if (__io_alloc_async_data(req))
Jens Axboeb76da702019-11-20 13:05:32 -07005795 return -EAGAIN;
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005796 return io_req_prep_async(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005797}
5798
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005799static u32 io_get_sequence(struct io_kiocb *req)
5800{
5801 struct io_kiocb *pos;
5802 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005803 u32 total_submitted, nr_reqs = 0;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005804
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005805 io_for_each_link(pos, req)
5806 nr_reqs++;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005807
5808 total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped;
5809 return total_submitted - nr_reqs;
5810}
5811
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005812static int io_req_defer(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07005813{
5814 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005815 struct io_defer_entry *de;
Jens Axboe2b188cc2019-01-07 10:46:33 -07005816 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005817 u32 seq;
Jens Axboe2b188cc2019-01-07 10:46:33 -07005818
5819 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005820 if (likely(list_empty_careful(&ctx->defer_list) &&
5821 !(req->flags & REQ_F_IO_DRAIN)))
5822 return 0;
5823
5824 seq = io_get_sequence(req);
5825 /* Still a chance to pass the sequence check */
5826 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Jens Axboe2b188cc2019-01-07 10:46:33 -07005827 return 0;
5828
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005829 ret = io_req_defer_prep(req);
5830 if (ret)
5831 return ret;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03005832 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005833 de = kmalloc(sizeof(*de), GFP_KERNEL);
5834 if (!de)
5835 return -ENOMEM;
Jens Axboe31b51512019-01-18 22:56:34 -07005836
5837 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005838 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07005839 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005840 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03005841 io_queue_async_work(req);
5842 return -EIOCBQUEUED;
Jens Axboe31b51512019-01-18 22:56:34 -07005843 }
5844
5845 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005846 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005847 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005848 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07005849 spin_unlock_irq(&ctx->completion_lock);
5850 return -EIOCBQUEUED;
5851}
Jens Axboeedafcce2019-01-09 09:16:05 -07005852
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03005853static void __io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005854{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005855 if (req->flags & REQ_F_BUFFER_SELECTED) {
5856 switch (req->opcode) {
5857 case IORING_OP_READV:
5858 case IORING_OP_READ_FIXED:
5859 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07005860 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005861 break;
5862 case IORING_OP_RECVMSG:
5863 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07005864 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005865 break;
5866 }
5867 req->flags &= ~REQ_F_BUFFER_SELECTED;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005868 }
5869
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005870 if (req->flags & REQ_F_NEED_CLEANUP) {
5871 switch (req->opcode) {
5872 case IORING_OP_READV:
5873 case IORING_OP_READ_FIXED:
5874 case IORING_OP_READ:
5875 case IORING_OP_WRITEV:
5876 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005877 case IORING_OP_WRITE: {
5878 struct io_async_rw *io = req->async_data;
5879 if (io->free_iovec)
5880 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005881 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005882 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005883 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005884 case IORING_OP_SENDMSG: {
5885 struct io_async_msghdr *io = req->async_data;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00005886
5887 kfree(io->free_iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005888 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005889 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005890 case IORING_OP_SPLICE:
5891 case IORING_OP_TEE:
5892 io_put_file(req, req->splice.file_in,
5893 (req->splice.flags & SPLICE_F_FD_IN_FIXED));
5894 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06005895 case IORING_OP_OPENAT:
5896 case IORING_OP_OPENAT2:
5897 if (req->open.filename)
5898 putname(req->open.filename);
5899 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06005900 case IORING_OP_RENAMEAT:
5901 putname(req->rename.oldpath);
5902 putname(req->rename.newpath);
5903 break;
Jens Axboe14a11432020-09-28 14:27:37 -06005904 case IORING_OP_UNLINKAT:
5905 putname(req->unlink.filename);
5906 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005907 }
5908 req->flags &= ~REQ_F_NEED_CLEANUP;
5909 }
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005910}
5911
Pavel Begunkov889fca72021-02-10 00:03:09 +00005912static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeedafcce2019-01-09 09:16:05 -07005913{
Jens Axboeedafcce2019-01-09 09:16:05 -07005914 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5730b272021-02-27 15:57:30 -07005915 const struct cred *creds = NULL;
Jens Axboed625c6e2019-12-17 19:53:05 -07005916 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07005917
Jens Axboe5730b272021-02-27 15:57:30 -07005918 if (req->work.personality) {
5919 const struct cred *new_creds;
5920
5921 if (!(issue_flags & IO_URING_F_NONBLOCK))
5922 mutex_lock(&ctx->uring_lock);
5923 new_creds = idr_find(&ctx->personality_idr, req->work.personality);
5924 if (!(issue_flags & IO_URING_F_NONBLOCK))
5925 mutex_unlock(&ctx->uring_lock);
5926 if (!new_creds)
5927 return -EINVAL;
5928 creds = override_creds(new_creds);
5929 }
5930
Jens Axboed625c6e2019-12-17 19:53:05 -07005931 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07005932 case IORING_OP_NOP:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005933 ret = io_nop(req, issue_flags);
Jens Axboe31b51512019-01-18 22:56:34 -07005934 break;
5935 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07005936 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005937 case IORING_OP_READ:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005938 ret = io_read(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005939 break;
5940 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07005941 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005942 case IORING_OP_WRITE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005943 ret = io_write(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005944 break;
5945 case IORING_OP_FSYNC:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005946 ret = io_fsync(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005947 break;
5948 case IORING_OP_POLL_ADD:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005949 ret = io_poll_add(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005950 break;
5951 case IORING_OP_POLL_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005952 ret = io_poll_remove(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005953 break;
5954 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005955 ret = io_sync_file_range(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005956 break;
5957 case IORING_OP_SENDMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005958 ret = io_sendmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005959 break;
Jens Axboefddafac2020-01-04 20:19:44 -07005960 case IORING_OP_SEND:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005961 ret = io_send(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07005962 break;
5963 case IORING_OP_RECVMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005964 ret = io_recvmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005965 break;
Jens Axboefddafac2020-01-04 20:19:44 -07005966 case IORING_OP_RECV:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005967 ret = io_recv(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005968 break;
5969 case IORING_OP_TIMEOUT:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005970 ret = io_timeout(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005971 break;
5972 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005973 ret = io_timeout_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005974 break;
5975 case IORING_OP_ACCEPT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005976 ret = io_accept(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005977 break;
5978 case IORING_OP_CONNECT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005979 ret = io_connect(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005980 break;
5981 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov61e98202021-02-10 00:03:08 +00005982 ret = io_async_cancel(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005983 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07005984 case IORING_OP_FALLOCATE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005985 ret = io_fallocate(req, issue_flags);
Jens Axboed63d1b52019-12-10 10:38:56 -07005986 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07005987 case IORING_OP_OPENAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005988 ret = io_openat(req, issue_flags);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005989 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07005990 case IORING_OP_CLOSE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005991 ret = io_close(req, issue_flags);
Jens Axboeb5dba592019-12-11 14:02:38 -07005992 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005993 case IORING_OP_FILES_UPDATE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00005994 ret = io_files_update(req, issue_flags);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005995 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005996 case IORING_OP_STATX:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005997 ret = io_statx(req, issue_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005998 break;
Jens Axboe4840e412019-12-25 22:03:45 -07005999 case IORING_OP_FADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006000 ret = io_fadvise(req, issue_flags);
Jens Axboe4840e412019-12-25 22:03:45 -07006001 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07006002 case IORING_OP_MADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006003 ret = io_madvise(req, issue_flags);
Jens Axboec1ca7572019-12-25 22:18:28 -07006004 break;
Jens Axboecebdb982020-01-08 17:59:24 -07006005 case IORING_OP_OPENAT2:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006006 ret = io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07006007 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07006008 case IORING_OP_EPOLL_CTL:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006009 ret = io_epoll_ctl(req, issue_flags);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006010 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006011 case IORING_OP_SPLICE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006012 ret = io_splice(req, issue_flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006013 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07006014 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006015 ret = io_provide_buffers(req, issue_flags);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006016 break;
Jens Axboe067524e2020-03-02 16:32:28 -07006017 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006018 ret = io_remove_buffers(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006019 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006020 case IORING_OP_TEE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006021 ret = io_tee(req, issue_flags);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006022 break;
Jens Axboe36f4fa62020-09-05 11:14:22 -06006023 case IORING_OP_SHUTDOWN:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006024 ret = io_shutdown(req, issue_flags);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006025 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006026 case IORING_OP_RENAMEAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006027 ret = io_renameat(req, issue_flags);
Jens Axboe80a261f2020-09-28 14:23:58 -06006028 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006029 case IORING_OP_UNLINKAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006030 ret = io_unlinkat(req, issue_flags);
Jens Axboe14a11432020-09-28 14:27:37 -06006031 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006032 default:
6033 ret = -EINVAL;
6034 break;
Jens Axboe31b51512019-01-18 22:56:34 -07006035 }
6036
Jens Axboe5730b272021-02-27 15:57:30 -07006037 if (creds)
6038 revert_creds(creds);
6039
Jens Axboe2b188cc2019-01-07 10:46:33 -07006040 if (ret)
6041 return ret;
6042
Jens Axboeb5325762020-05-19 21:20:27 -06006043 /* If the op doesn't have a file, we're not polling for it */
6044 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
Jens Axboe11ba8202020-01-15 21:51:17 -07006045 const bool in_async = io_wq_current_is_worker();
6046
Jens Axboe11ba8202020-01-15 21:51:17 -07006047 /* workqueue context doesn't hold uring_lock, grab it now */
6048 if (in_async)
6049 mutex_lock(&ctx->uring_lock);
6050
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08006051 io_iopoll_req_issued(req, in_async);
Jens Axboe11ba8202020-01-15 21:51:17 -07006052
6053 if (in_async)
6054 mutex_unlock(&ctx->uring_lock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006055 }
6056
6057 return 0;
6058}
6059
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00006060static void io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006061{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006062 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006063 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006064 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006065
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006066 timeout = io_prep_linked_timeout(req);
6067 if (timeout)
6068 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006069
Jens Axboe4014d942021-01-19 15:53:54 -07006070 if (work->flags & IO_WQ_WORK_CANCEL)
Jens Axboe561fb042019-10-24 07:25:42 -06006071 ret = -ECANCELED;
Jens Axboe31b51512019-01-18 22:56:34 -07006072
Jens Axboe561fb042019-10-24 07:25:42 -06006073 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006074 do {
Pavel Begunkov889fca72021-02-10 00:03:09 +00006075 ret = io_issue_sqe(req, 0);
Jens Axboe561fb042019-10-24 07:25:42 -06006076 /*
6077 * We can get EAGAIN for polled IO even though we're
6078 * forcing a sync submission from here, since we can't
6079 * wait for request slots on the block side.
6080 */
6081 if (ret != -EAGAIN)
6082 break;
6083 cond_resched();
6084 } while (1);
6085 }
Jens Axboe31b51512019-01-18 22:56:34 -07006086
Pavel Begunkova3df76982021-02-18 22:32:52 +00006087 /* avoid locking problems by failing it from a clean context */
Jens Axboe561fb042019-10-24 07:25:42 -06006088 if (ret) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00006089 /* io-wq is going to take one down */
6090 refcount_inc(&req->refs);
6091 io_req_task_queue_fail(req, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07006092 }
Jens Axboe31b51512019-01-18 22:56:34 -07006093}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006094
Jens Axboe65e19f52019-10-26 07:20:21 -06006095static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6096 int index)
Jens Axboe09bb8392019-03-13 12:39:28 -06006097{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006098 struct fixed_rsrc_table *table;
Jens Axboe65e19f52019-10-26 07:20:21 -06006099
Jens Axboe05f3fb32019-12-09 11:22:50 -07006100 table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT];
Xiaoming Ni84695082020-05-11 19:25:43 +08006101 return table->files[index & IORING_FILE_TABLE_MASK];
Jens Axboe65e19f52019-10-26 07:20:21 -06006102}
6103
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006104static struct file *io_file_get(struct io_submit_state *state,
6105 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006106{
6107 struct io_ring_ctx *ctx = req->ctx;
6108 struct file *file;
6109
6110 if (fixed) {
Pavel Begunkov479f5172020-10-10 18:34:07 +01006111 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006112 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006113 fd = array_index_nospec(fd, ctx->nr_user_files);
6114 file = io_file_from_index(ctx, fd);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00006115 io_set_resource_node(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006116 } else {
6117 trace_io_uring_file_get(ctx, fd);
6118 file = __io_file_get(state, fd);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006119 }
6120
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00006121 if (file && unlikely(file->f_op == &io_uring_fops))
6122 io_req_track_inflight(req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006123 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006124}
6125
Jens Axboe2665abf2019-11-05 12:40:47 -07006126static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6127{
Jens Axboead8a48a2019-11-15 08:49:11 -07006128 struct io_timeout_data *data = container_of(timer,
6129 struct io_timeout_data, timer);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006130 struct io_kiocb *prev, *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006131 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2665abf2019-11-05 12:40:47 -07006132 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006133
6134 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006135 prev = req->timeout.head;
6136 req->timeout.head = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006137
6138 /*
6139 * We don't expect the list to be empty, that will only happen if we
6140 * race with the completion of the linked work.
6141 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006142 if (prev && refcount_inc_not_zero(&prev->refs))
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006143 io_remove_next_linked(prev);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006144 else
6145 prev = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006146 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6147
6148 if (prev) {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006149 req_set_fail_links(prev);
Pavel Begunkov014db002020-03-03 21:33:12 +03006150 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006151 io_put_req_deferred(prev, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07006152 } else {
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006153 io_req_complete_post(req, -ETIME, 0);
6154 io_put_req_deferred(req, 1);
Jens Axboe2665abf2019-11-05 12:40:47 -07006155 }
Jens Axboe2665abf2019-11-05 12:40:47 -07006156 return HRTIMER_NORESTART;
6157}
6158
Jens Axboe7271ef32020-08-10 09:55:22 -06006159static void __io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006160{
Jens Axboe76a46e02019-11-10 23:34:16 -07006161 /*
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006162 * If the back reference is NULL, then our linked request finished
6163 * before we got a chance to setup the timer
Jens Axboe76a46e02019-11-10 23:34:16 -07006164 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006165 if (req->timeout.head) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006166 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006167
Jens Axboead8a48a2019-11-15 08:49:11 -07006168 data->timer.function = io_link_timeout_fn;
6169 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6170 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006171 }
Jens Axboe7271ef32020-08-10 09:55:22 -06006172}
6173
6174static void io_queue_linked_timeout(struct io_kiocb *req)
6175{
6176 struct io_ring_ctx *ctx = req->ctx;
6177
6178 spin_lock_irq(&ctx->completion_lock);
6179 __io_queue_linked_timeout(req);
Jens Axboe76a46e02019-11-10 23:34:16 -07006180 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006181
Jens Axboe2665abf2019-11-05 12:40:47 -07006182 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006183 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006184}
6185
Jens Axboead8a48a2019-11-15 08:49:11 -07006186static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006187{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006188 struct io_kiocb *nxt = req->link;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006189
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006190 if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) ||
6191 nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006192 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006193
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006194 nxt->timeout.head = req;
Pavel Begunkov900fad42020-10-19 16:39:16 +01006195 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006196 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006197 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006198}
6199
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006200static void __io_queue_sqe(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006201{
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006202 struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006203 int ret;
6204
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006205 ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
Jens Axboe491381ce2019-10-17 09:20:46 -06006206
6207 /*
6208 * We async punt it if the file wasn't marked NOWAIT, or if the file
6209 * doesn't support non-blocking read/write attempts
6210 */
Pavel Begunkov24c74672020-06-21 13:09:51 +03006211 if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006212 if (!io_arm_poll_handler(req)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006213 /*
6214 * Queued up for async execution, worker will release
6215 * submit reference when the iocb is actually submitted.
6216 */
6217 io_queue_async_work(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006218 }
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006219 } else if (likely(!ret)) {
6220 /* drop submission reference */
Pavel Begunkove342c802021-01-19 13:32:47 +00006221 if (req->flags & REQ_F_COMPLETE_INLINE) {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006222 struct io_ring_ctx *ctx = req->ctx;
6223 struct io_comp_state *cs = &ctx->submit_state.comp;
Jens Axboee65ef562019-03-12 10:16:44 -06006224
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006225 cs->reqs[cs->nr++] = req;
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006226 if (cs->nr == ARRAY_SIZE(cs->reqs))
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006227 io_submit_flush_completions(cs, ctx);
Pavel Begunkov9affd662021-01-19 13:32:46 +00006228 } else {
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006229 io_put_req(req);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006230 }
6231 } else {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006232 req_set_fail_links(req);
Jens Axboee65ef562019-03-12 10:16:44 -06006233 io_put_req(req);
Pavel Begunkov652532a2020-07-03 22:15:07 +03006234 io_req_complete(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006235 }
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006236 if (linked_timeout)
6237 io_queue_linked_timeout(linked_timeout);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006238}
6239
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006240static void io_queue_sqe(struct io_kiocb *req)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006241{
6242 int ret;
6243
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006244 ret = io_req_defer(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006245 if (ret) {
6246 if (ret != -EIOCBQUEUED) {
Pavel Begunkov11185912020-01-22 23:09:35 +03006247fail_req:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006248 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06006249 io_put_req(req);
6250 io_req_complete(req, ret);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006251 }
Pavel Begunkov25508782019-12-30 21:24:47 +03006252 } else if (req->flags & REQ_F_FORCE_ASYNC) {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006253 ret = io_req_defer_prep(req);
6254 if (unlikely(ret))
6255 goto fail_req;
Jens Axboece35a472019-12-17 08:04:44 -07006256 io_queue_async_work(req);
6257 } else {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006258 __io_queue_sqe(req);
Jens Axboece35a472019-12-17 08:04:44 -07006259 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006260}
6261
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006262/*
6263 * Check SQE restrictions (opcode and flags).
6264 *
6265 * Returns 'true' if SQE is allowed, 'false' otherwise.
6266 */
6267static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6268 struct io_kiocb *req,
6269 unsigned int sqe_flags)
6270{
6271 if (!ctx->restricted)
6272 return true;
6273
6274 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6275 return false;
6276
6277 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6278 ctx->restrictions.sqe_flags_required)
6279 return false;
6280
6281 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6282 ctx->restrictions.sqe_flags_required))
6283 return false;
6284
6285 return true;
6286}
6287
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006288static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006289 const struct io_uring_sqe *sqe)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006290{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006291 struct io_submit_state *state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006292 unsigned int sqe_flags;
Jens Axboe5730b272021-02-27 15:57:30 -07006293 int ret = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006294
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006295 req->opcode = READ_ONCE(sqe->opcode);
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006296 /* same numerical values with corresponding REQ_F_*, safe to copy */
6297 req->flags = sqe_flags = READ_ONCE(sqe->flags);
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006298 req->user_data = READ_ONCE(sqe->user_data);
Jens Axboee8c2bc12020-08-15 18:44:09 -07006299 req->async_data = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006300 req->file = NULL;
6301 req->ctx = ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006302 req->link = NULL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006303 req->fixed_rsrc_refs = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006304 /* one is dropped after submission, the other at completion */
6305 refcount_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006306 req->task = current;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006307 req->result = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006308
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006309 /* enforce forwards compatibility on users */
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006310 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) {
6311 req->flags = 0;
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006312 return -EINVAL;
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006313 }
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006314
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006315 if (unlikely(req->opcode >= IORING_OP_LAST))
6316 return -EINVAL;
6317
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006318 if (unlikely(!io_check_restriction(ctx, req, sqe_flags)))
6319 return -EACCES;
6320
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006321 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6322 !io_op_defs[req->opcode].buffer_select)
6323 return -EOPNOTSUPP;
6324
Jens Axboe5730b272021-02-27 15:57:30 -07006325 req->work.list.next = NULL;
6326 req->work.flags = 0;
6327 req->work.personality = READ_ONCE(sqe->personality);
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006328 state = &ctx->submit_state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006329
Jens Axboe27926b62020-10-28 09:33:23 -06006330 /*
6331 * Plug now if we have more than 1 IO left after this, and the target
6332 * is potentially a read/write to block based storage.
6333 */
6334 if (!state->plug_started && state->ios_left > 1 &&
6335 io_op_defs[req->opcode].plug) {
6336 blk_start_plug(&state->plug);
6337 state->plug_started = true;
6338 }
Jens Axboe63ff8222020-05-07 14:56:15 -06006339
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006340 if (io_op_defs[req->opcode].needs_file) {
6341 bool fixed = req->flags & REQ_F_FIXED_FILE;
Jens Axboe63ff8222020-05-07 14:56:15 -06006342
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006343 req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed);
Pavel Begunkovba13e232021-02-01 18:59:52 +00006344 if (unlikely(!req->file))
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006345 ret = -EBADF;
6346 }
6347
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006348 state->ios_left--;
6349 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006350}
6351
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006352static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006353 const struct io_uring_sqe *sqe)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006354{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006355 struct io_submit_link *link = &ctx->submit_state.link;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006356 int ret;
6357
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006358 ret = io_init_req(ctx, req, sqe);
6359 if (unlikely(ret)) {
6360fail_req:
6361 io_put_req(req);
6362 io_req_complete(req, ret);
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006363 if (link->head) {
6364 /* fail even hard links since we don't submit */
Pavel Begunkovcf109602021-02-18 18:29:43 +00006365 link->head->flags |= REQ_F_FAIL_LINK;
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006366 io_put_req(link->head);
6367 io_req_complete(link->head, -ECANCELED);
6368 link->head = NULL;
6369 }
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006370 return ret;
6371 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006372 ret = io_req_prep(req, sqe);
6373 if (unlikely(ret))
6374 goto fail_req;
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006375
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006376 /* don't need @sqe from now on */
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006377 trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data,
6378 true, ctx->flags & IORING_SETUP_SQPOLL);
6379
Jens Axboe6c271ce2019-01-10 11:22:30 -07006380 /*
6381 * If we already have a head request, queue this one for async
6382 * submittal once the head completes. If we don't have a head but
6383 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6384 * submitted sync once the chain is complete. If none of those
6385 * conditions are true (normal request), then just queue it.
6386 */
6387 if (link->head) {
6388 struct io_kiocb *head = link->head;
6389
6390 /*
6391 * Taking sequential execution of a link, draining both sides
6392 * of the link also fullfils IOSQE_IO_DRAIN semantics for all
6393 * requests in the link. So, it drains the head and the
6394 * next after the link request. The last one is done via
6395 * drain_next flag to persist the effect across calls.
6396 */
6397 if (req->flags & REQ_F_IO_DRAIN) {
6398 head->flags |= REQ_F_IO_DRAIN;
6399 ctx->drain_next = 1;
6400 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006401 ret = io_req_defer_prep(req);
Pavel Begunkovcf109602021-02-18 18:29:43 +00006402 if (unlikely(ret))
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006403 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006404 trace_io_uring_link(ctx, req, head);
6405 link->last->link = req;
6406 link->last = req;
6407
6408 /* last request of a link, enqueue the link */
6409 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006410 io_queue_sqe(head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006411 link->head = NULL;
6412 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006413 } else {
6414 if (unlikely(ctx->drain_next)) {
6415 req->flags |= REQ_F_IO_DRAIN;
6416 ctx->drain_next = 0;
6417 }
6418 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Jackie Liu4fe2c962019-09-09 20:50:40 +08006419 link->head = req;
6420 link->last = req;
6421 } else {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006422 io_queue_sqe(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006423 }
6424 }
6425
6426 return 0;
6427}
6428
6429/*
6430 * Batched submission is done, ensure local IO is flushed out.
6431 */
6432static void io_submit_state_end(struct io_submit_state *state,
6433 struct io_ring_ctx *ctx)
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006434{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006435 if (state->link.head)
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006436 io_queue_sqe(state->link.head);
Jens Axboe3529d8c2019-12-19 18:24:38 -07006437 if (state->comp.nr)
Jens Axboe9e645e112019-05-10 16:07:28 -06006438 io_submit_flush_completions(&state->comp, ctx);
Jackie Liua197f662019-11-08 08:09:12 -07006439 if (state->plug_started)
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006440 blk_finish_plug(&state->plug);
Jens Axboe75c6a032020-01-28 10:15:23 -07006441 io_state_file_put(state);
Jens Axboe9e645e112019-05-10 16:07:28 -06006442}
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006443
Jens Axboe9e645e112019-05-10 16:07:28 -06006444/*
6445 * Start submission side cache.
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006446 */
Jens Axboe9e645e112019-05-10 16:07:28 -06006447static void io_submit_state_start(struct io_submit_state *state,
Pavel Begunkov196be952019-11-07 01:41:06 +03006448 unsigned int max_ios)
Jens Axboe9e645e112019-05-10 16:07:28 -06006449{
6450 state->plug_started = false;
Jens Axboebcda7ba2020-02-23 16:42:51 -07006451 state->ios_left = max_ios;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006452 /* set only head, no need to init link_last in advance */
6453 state->link.head = NULL;
Jens Axboe75c6a032020-01-28 10:15:23 -07006454}
6455
Jens Axboe193155c2020-02-22 23:22:19 -07006456static void io_commit_sqring(struct io_ring_ctx *ctx)
6457{
Jens Axboe75c6a032020-01-28 10:15:23 -07006458 struct io_rings *rings = ctx->rings;
6459
6460 /*
Jens Axboe193155c2020-02-22 23:22:19 -07006461 * Ensure any loads from the SQEs are done at this point,
Jens Axboe75c6a032020-01-28 10:15:23 -07006462 * since once we write the new head, the application could
6463 * write new data to them.
Pavel Begunkov6b47ee62020-01-18 20:22:41 +03006464 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006465 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboebcda7ba2020-02-23 16:42:51 -07006466}
6467
Jens Axboe9e645e112019-05-10 16:07:28 -06006468/*
Jens Axboe3529d8c2019-12-19 18:24:38 -07006469 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
Jens Axboe9e645e112019-05-10 16:07:28 -06006470 * that is mapped by userspace. This means that care needs to be taken to
6471 * ensure that reads are stable, as we cannot rely on userspace always
Jens Axboe78e19bb2019-11-06 15:21:34 -07006472 * being a good citizen. If members of the sqe are validated and then later
6473 * used, it's important that those reads are done through READ_ONCE() to
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006474 * prevent a re-load down the line.
Jens Axboe9e645e112019-05-10 16:07:28 -06006475 */
6476static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe9e645e112019-05-10 16:07:28 -06006477{
6478 u32 *sq_array = ctx->sq_array;
6479 unsigned head;
6480
6481 /*
6482 * The cached sq head (or cq tail) serves two purposes:
6483 *
6484 * 1) allows us to batch the cost of updating the user visible
Pavel Begunkov9d763772019-12-17 02:22:07 +03006485 * head updates.
Jens Axboe9e645e112019-05-10 16:07:28 -06006486 * 2) allows the kernel side to track the head on its own, even
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006487 * though the application is the one updating it.
6488 */
6489 head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]);
6490 if (likely(head < ctx->sq_entries))
6491 return &ctx->sq_sqes[head];
6492
6493 /* drop invalid entries */
Pavel Begunkov711be032020-01-17 03:57:59 +03006494 ctx->cached_sq_dropped++;
6495 WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped);
6496 return NULL;
6497}
Jens Axboeb7bb4f72019-12-15 22:13:43 -07006498
Jens Axboe0f212202020-09-13 13:09:39 -06006499static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006500{
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006501 int submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006502
Jens Axboec4a2ed72019-11-21 21:01:26 -07006503 /* if we have a backlog and couldn't flush it all, return BUSY */
Jens Axboead3eb2c2019-12-18 17:12:20 -07006504 if (test_bit(0, &ctx->sq_check_overflow)) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006505 if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL))
Jens Axboead3eb2c2019-12-18 17:12:20 -07006506 return -EBUSY;
6507 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006508
Pavel Begunkovee7d46d2019-12-30 21:24:45 +03006509 /* make sure SQ entry isn't read before tail */
6510 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov9ef4f122019-12-30 21:24:44 +03006511
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006512 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6513 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006514
Jens Axboed8a6df12020-10-15 16:24:45 -06006515 percpu_counter_add(&current->io_uring->inflight, nr);
Jens Axboefaf7b512020-10-07 12:48:53 -06006516 refcount_add(nr, &current->usage);
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006517 io_submit_state_start(&ctx->submit_state, nr);
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006518
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006519 while (submitted < nr) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006520 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006521 struct io_kiocb *req;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006522
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006523 req = io_alloc_req(ctx);
Pavel Begunkov196be952019-11-07 01:41:06 +03006524 if (unlikely(!req)) {
6525 if (!submitted)
6526 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006527 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006528 }
Pavel Begunkov4fccfcb2021-02-12 11:55:17 +00006529 sqe = io_get_sqe(ctx);
6530 if (unlikely(!sqe)) {
6531 kmem_cache_free(req_cachep, req);
6532 break;
6533 }
Jens Axboed3656342019-12-18 09:50:26 -07006534 /* will complete beyond this point, count as submitted */
6535 submitted++;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006536 if (io_submit_sqe(ctx, req, sqe))
Jens Axboed3656342019-12-18 09:50:26 -07006537 break;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006538 }
6539
Pavel Begunkov9466f432020-01-25 22:34:01 +03006540 if (unlikely(submitted != nr)) {
6541 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006542 struct io_uring_task *tctx = current->io_uring;
6543 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006544
Jens Axboed8a6df12020-10-15 16:24:45 -06006545 percpu_ref_put_many(&ctx->refs, unused);
6546 percpu_counter_sub(&tctx->inflight, unused);
6547 put_task_struct_many(current, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006548 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006549
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006550 io_submit_state_end(&ctx->submit_state, ctx);
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006551 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6552 io_commit_sqring(ctx);
6553
Jens Axboe6c271ce2019-01-10 11:22:30 -07006554 return submitted;
6555}
6556
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006557static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6558{
6559 /* Tell userspace we may need a wakeup call */
6560 spin_lock_irq(&ctx->completion_lock);
6561 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6562 spin_unlock_irq(&ctx->completion_lock);
6563}
6564
6565static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6566{
6567 spin_lock_irq(&ctx->completion_lock);
6568 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6569 spin_unlock_irq(&ctx->completion_lock);
6570}
6571
Xiaoguang Wang08369242020-11-03 14:15:59 +08006572static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006573{
Jens Axboec8d1ba52020-09-14 11:07:26 -06006574 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006575 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006576
Jens Axboec8d1ba52020-09-14 11:07:26 -06006577 to_submit = io_sqring_entries(ctx);
Jens Axboee95eee22020-09-08 09:11:32 -06006578 /* if we're handling multiple rings, cap submit size for fairness */
6579 if (cap_entries && to_submit > 8)
6580 to_submit = 8;
6581
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006582 if (!list_empty(&ctx->iopoll_list) || to_submit) {
6583 unsigned nr_events = 0;
6584
Xiaoguang Wang08369242020-11-03 14:15:59 +08006585 mutex_lock(&ctx->uring_lock);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006586 if (!list_empty(&ctx->iopoll_list))
6587 io_do_iopoll(ctx, &nr_events, 0);
6588
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00006589 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)))
Xiaoguang Wang08369242020-11-03 14:15:59 +08006590 ret = io_submit_sqes(ctx, to_submit);
6591 mutex_unlock(&ctx->uring_lock);
6592 }
Jens Axboe90554202020-09-03 12:12:41 -06006593
6594 if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait))
6595 wake_up(&ctx->sqo_sq_wait);
6596
Xiaoguang Wang08369242020-11-03 14:15:59 +08006597 return ret;
6598}
6599
6600static void io_sqd_update_thread_idle(struct io_sq_data *sqd)
6601{
6602 struct io_ring_ctx *ctx;
6603 unsigned sq_thread_idle = 0;
6604
6605 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6606 if (sq_thread_idle < ctx->sq_thread_idle)
6607 sq_thread_idle = ctx->sq_thread_idle;
6608 }
6609
6610 sqd->sq_thread_idle = sq_thread_idle;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006611}
6612
Jens Axboe69fb2132020-09-14 11:16:23 -06006613static void io_sqd_init_new(struct io_sq_data *sqd)
6614{
6615 struct io_ring_ctx *ctx;
6616
6617 while (!list_empty(&sqd->ctx_new_list)) {
6618 ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006619 list_move_tail(&ctx->sqd_list, &sqd->ctx_list);
6620 complete(&ctx->sq_thread_comp);
6621 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006622
6623 io_sqd_update_thread_idle(sqd);
Jens Axboe69fb2132020-09-14 11:16:23 -06006624}
6625
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006626static bool io_sq_thread_should_stop(struct io_sq_data *sqd)
6627{
6628 return test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
6629}
6630
6631static bool io_sq_thread_should_park(struct io_sq_data *sqd)
6632{
6633 return test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
6634}
6635
6636static void io_sq_thread_parkme(struct io_sq_data *sqd)
6637{
6638 for (;;) {
6639 /*
6640 * TASK_PARKED is a special state; we must serialize against
6641 * possible pending wakeups to avoid store-store collisions on
6642 * task->state.
6643 *
6644 * Such a collision might possibly result in the task state
6645 * changin from TASK_PARKED and us failing the
6646 * wait_task_inactive() in kthread_park().
6647 */
6648 set_special_state(TASK_PARKED);
6649 if (!test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state))
6650 break;
6651
6652 /*
6653 * Thread is going to call schedule(), do not preempt it,
6654 * or the caller of kthread_park() may spend more time in
6655 * wait_task_inactive().
6656 */
6657 preempt_disable();
6658 complete(&sqd->completion);
6659 schedule_preempt_disabled();
6660 preempt_enable();
6661 }
6662 __set_current_state(TASK_RUNNING);
6663}
6664
Jens Axboe6c271ce2019-01-10 11:22:30 -07006665static int io_sq_thread(void *data)
6666{
Jens Axboe69fb2132020-09-14 11:16:23 -06006667 struct io_sq_data *sqd = data;
6668 struct io_ring_ctx *ctx;
Xiaoguang Wanga0d92052020-11-12 14:55:59 +08006669 unsigned long timeout = 0;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006670 char buf[TASK_COMM_LEN];
Xiaoguang Wang08369242020-11-03 14:15:59 +08006671 DEFINE_WAIT(wait);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006672
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006673 sprintf(buf, "iou-sqp-%d", sqd->task_pid);
6674 set_task_comm(current, buf);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006675 current->pf_io_worker = NULL;
Jens Axboe28cea78a2020-09-14 10:51:17 -06006676
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006677 if (sqd->sq_cpu != -1)
6678 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
6679 else
6680 set_cpus_allowed_ptr(current, cpu_online_mask);
6681 current->flags |= PF_NO_SETAFFINITY;
6682
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006683 wait_for_completion(&sqd->startup);
6684
6685 while (!io_sq_thread_should_stop(sqd)) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08006686 int ret;
6687 bool cap_entries, sqt_spin, needs_sched;
Jens Axboec1edbf52019-11-10 16:56:04 -07006688
6689 /*
Jens Axboe69fb2132020-09-14 11:16:23 -06006690 * Any changes to the sqd lists are synchronized through the
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006691 * thread parking. This synchronizes the thread vs users,
Jens Axboe69fb2132020-09-14 11:16:23 -06006692 * the users are synchronized on the sqd->ctx_lock.
Jens Axboec1edbf52019-11-10 16:56:04 -07006693 */
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006694 if (io_sq_thread_should_park(sqd)) {
6695 io_sq_thread_parkme(sqd);
6696 continue;
Xiaoguang Wang65b2b212020-11-19 17:44:46 +08006697 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006698 if (unlikely(!list_empty(&sqd->ctx_new_list))) {
Jens Axboe69fb2132020-09-14 11:16:23 -06006699 io_sqd_init_new(sqd);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006700 timeout = jiffies + sqd->sq_thread_idle;
6701 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006702 if (fatal_signal_pending(current))
6703 break;
Xiaoguang Wang08369242020-11-03 14:15:59 +08006704 sqt_spin = false;
Jens Axboee95eee22020-09-08 09:11:32 -06006705 cap_entries = !list_is_singular(&sqd->ctx_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006706 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08006707 ret = __io_sq_thread(ctx, cap_entries);
6708 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
6709 sqt_spin = true;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006710 }
6711
Xiaoguang Wang08369242020-11-03 14:15:59 +08006712 if (sqt_spin || !time_after(jiffies, timeout)) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06006713 io_run_task_work();
6714 cond_resched();
Xiaoguang Wang08369242020-11-03 14:15:59 +08006715 if (sqt_spin)
6716 timeout = jiffies + sqd->sq_thread_idle;
6717 continue;
6718 }
6719
Xiaoguang Wang08369242020-11-03 14:15:59 +08006720 needs_sched = true;
6721 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
6722 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6723 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
6724 !list_empty_careful(&ctx->iopoll_list)) {
6725 needs_sched = false;
6726 break;
6727 }
6728 if (io_sqring_entries(ctx)) {
6729 needs_sched = false;
6730 break;
6731 }
6732 }
6733
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006734 if (needs_sched && !io_sq_thread_should_park(sqd)) {
Jens Axboe69fb2132020-09-14 11:16:23 -06006735 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6736 io_ring_set_wakeup_flag(ctx);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006737
Jens Axboe69fb2132020-09-14 11:16:23 -06006738 schedule();
Jens Axboee4b4a132021-03-01 18:36:25 -07006739 try_to_freeze();
Jens Axboe69fb2132020-09-14 11:16:23 -06006740 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6741 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006742 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006743
6744 finish_wait(&sqd->wait, &wait);
6745 timeout = jiffies + sqd->sq_thread_idle;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006746 }
6747
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006748 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6749 io_uring_cancel_sqpoll(ctx);
6750
Jens Axboe4c6e2772020-07-01 11:29:10 -06006751 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07006752
Jens Axboe86293972021-02-26 13:46:49 -07006753 if (io_sq_thread_should_park(sqd))
6754 io_sq_thread_parkme(sqd);
6755
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006756 /*
6757 * Clear thread under lock so that concurrent parks work correctly
6758 */
Jens Axboe86293972021-02-26 13:46:49 -07006759 complete(&sqd->completion);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006760 mutex_lock(&sqd->lock);
6761 sqd->thread = NULL;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07006762 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6763 ctx->sqo_exec = 1;
6764 io_ring_set_wakeup_flag(ctx);
6765 }
Jens Axboe06058632019-04-13 09:26:03 -06006766
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006767 complete(&sqd->exited);
Jens Axboee54945a2021-02-26 11:27:15 -07006768 mutex_unlock(&sqd->lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006769 do_exit(0);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006770}
6771
Jens Axboebda52162019-09-24 13:47:15 -06006772struct io_wait_queue {
6773 struct wait_queue_entry wq;
6774 struct io_ring_ctx *ctx;
6775 unsigned to_wait;
6776 unsigned nr_timeouts;
6777};
6778
Pavel Begunkov6c503152021-01-04 20:36:36 +00006779static inline bool io_should_wake(struct io_wait_queue *iowq)
Jens Axboebda52162019-09-24 13:47:15 -06006780{
6781 struct io_ring_ctx *ctx = iowq->ctx;
6782
6783 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08006784 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06006785 * started waiting. For timeouts, we always want to return to userspace,
6786 * regardless of event count.
6787 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00006788 return io_cqring_events(ctx) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06006789 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
6790}
6791
6792static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
6793 int wake_flags, void *key)
6794{
6795 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
6796 wq);
6797
Pavel Begunkov6c503152021-01-04 20:36:36 +00006798 /*
6799 * Cannot safely flush overflowed CQEs from here, ensure we wake up
6800 * the task, and the next invocation will do it.
6801 */
6802 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
6803 return autoremove_wake_function(curr, mode, wake_flags, key);
6804 return -1;
Jens Axboebda52162019-09-24 13:47:15 -06006805}
6806
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006807static int io_run_task_work_sig(void)
6808{
6809 if (io_run_task_work())
6810 return 1;
6811 if (!signal_pending(current))
6812 return 0;
Jens Axboe792ee0f62020-10-22 20:17:18 -06006813 if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL))
6814 return -ERESTARTSYS;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006815 return -EINTR;
6816}
6817
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006818/* when returns >0, the caller should retry */
6819static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
6820 struct io_wait_queue *iowq,
6821 signed long *timeout)
6822{
6823 int ret;
6824
6825 /* make sure we run task_work before checking for signals */
6826 ret = io_run_task_work_sig();
6827 if (ret || io_should_wake(iowq))
6828 return ret;
6829 /* let the caller flush overflows, retry */
6830 if (test_bit(0, &ctx->cq_check_overflow))
6831 return 1;
6832
6833 *timeout = schedule_timeout(*timeout);
6834 return !*timeout ? -ETIME : 1;
6835}
6836
Jens Axboe2b188cc2019-01-07 10:46:33 -07006837/*
6838 * Wait until events become available, if we don't already have some. The
6839 * application must reap them itself, as they reside on the shared cq ring.
6840 */
6841static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
Hao Xuc73ebb62020-11-03 10:54:37 +08006842 const sigset_t __user *sig, size_t sigsz,
6843 struct __kernel_timespec __user *uts)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006844{
Jens Axboebda52162019-09-24 13:47:15 -06006845 struct io_wait_queue iowq = {
6846 .wq = {
6847 .private = current,
6848 .func = io_wake_function,
6849 .entry = LIST_HEAD_INIT(iowq.wq.entry),
6850 },
6851 .ctx = ctx,
6852 .to_wait = min_events,
6853 };
Hristo Venev75b28af2019-08-26 17:23:46 +00006854 struct io_rings *rings = ctx->rings;
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006855 signed long timeout = MAX_SCHEDULE_TIMEOUT;
6856 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006857
Jens Axboeb41e9852020-02-17 09:52:41 -07006858 do {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006859 io_cqring_overflow_flush(ctx, false, NULL, NULL);
6860 if (io_cqring_events(ctx) >= min_events)
Jens Axboeb41e9852020-02-17 09:52:41 -07006861 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06006862 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07006863 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07006864 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006865
6866 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006867#ifdef CONFIG_COMPAT
6868 if (in_compat_syscall())
6869 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07006870 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006871 else
6872#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07006873 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006874
Jens Axboe2b188cc2019-01-07 10:46:33 -07006875 if (ret)
6876 return ret;
6877 }
6878
Hao Xuc73ebb62020-11-03 10:54:37 +08006879 if (uts) {
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006880 struct timespec64 ts;
6881
Hao Xuc73ebb62020-11-03 10:54:37 +08006882 if (get_timespec64(&ts, uts))
6883 return -EFAULT;
6884 timeout = timespec64_to_jiffies(&ts);
6885 }
6886
Jens Axboebda52162019-09-24 13:47:15 -06006887 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02006888 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06006889 do {
Jens Axboeca0a2652021-03-04 17:15:48 -07006890 /* if we can't even flush overflow, don't wait for more */
6891 if (!io_cqring_overflow_flush(ctx, false, NULL, NULL)) {
6892 ret = -EBUSY;
6893 break;
6894 }
Jens Axboebda52162019-09-24 13:47:15 -06006895 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
6896 TASK_INTERRUPTIBLE);
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006897 ret = io_cqring_wait_schedule(ctx, &iowq, &timeout);
6898 finish_wait(&ctx->wait, &iowq.wq);
Jens Axboeca0a2652021-03-04 17:15:48 -07006899 cond_resched();
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006900 } while (ret > 0);
Jens Axboebda52162019-09-24 13:47:15 -06006901
Jens Axboeb7db41c2020-07-04 08:55:50 -06006902 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006903
Hristo Venev75b28af2019-08-26 17:23:46 +00006904 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006905}
6906
Jens Axboe6b063142019-01-10 22:13:58 -07006907static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
6908{
6909#if defined(CONFIG_UNIX)
6910 if (ctx->ring_sock) {
6911 struct sock *sock = ctx->ring_sock->sk;
6912 struct sk_buff *skb;
6913
6914 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
6915 kfree_skb(skb);
6916 }
6917#else
6918 int i;
6919
Jens Axboe65e19f52019-10-26 07:20:21 -06006920 for (i = 0; i < ctx->nr_user_files; i++) {
6921 struct file *file;
6922
6923 file = io_file_from_index(ctx, i);
6924 if (file)
6925 fput(file);
6926 }
Jens Axboe6b063142019-01-10 22:13:58 -07006927#endif
6928}
6929
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00006930static void io_rsrc_data_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07006931{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006932 struct fixed_rsrc_data *data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006933
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006934 data = container_of(ref, struct fixed_rsrc_data, refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006935 complete(&data->done);
6936}
6937
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006938static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
Pavel Begunkov1642b442020-12-30 21:34:14 +00006939{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006940 spin_lock_bh(&ctx->rsrc_ref_lock);
Pavel Begunkov1642b442020-12-30 21:34:14 +00006941}
6942
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006943static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx)
Jens Axboe6b063142019-01-10 22:13:58 -07006944{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006945 spin_unlock_bh(&ctx->rsrc_ref_lock);
6946}
6947
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00006948static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx,
6949 struct fixed_rsrc_data *rsrc_data,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006950 struct fixed_rsrc_ref_node *ref_node)
Jens Axboe6b063142019-01-10 22:13:58 -07006951{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006952 io_rsrc_ref_lock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006953 rsrc_data->node = ref_node;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00006954 list_add_tail(&ref_node->node, &ctx->rsrc_ref_list);
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006955 io_rsrc_ref_unlock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006956 percpu_ref_get(&rsrc_data->refs);
Jens Axboe6b063142019-01-10 22:13:58 -07006957}
6958
Hao Xu8bad28d2021-02-19 17:19:36 +08006959static void io_sqe_rsrc_kill_node(struct io_ring_ctx *ctx, struct fixed_rsrc_data *data)
Jens Axboe6b063142019-01-10 22:13:58 -07006960{
Hao Xu8bad28d2021-02-19 17:19:36 +08006961 struct fixed_rsrc_ref_node *ref_node = NULL;
Jens Axboe65e19f52019-10-26 07:20:21 -06006962
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006963 io_rsrc_ref_lock(ctx);
Pavel Begunkov1e5d7702020-11-18 14:56:25 +00006964 ref_node = data->node;
Pavel Begunkove6cb0072021-02-20 18:03:47 +00006965 data->node = NULL;
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00006966 io_rsrc_ref_unlock(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08006967 if (ref_node)
6968 percpu_ref_kill(&ref_node->refs);
Hao Xu8bad28d2021-02-19 17:19:36 +08006969}
Xiaoguang Wang05589552020-03-31 14:05:18 +08006970
Hao Xu8bad28d2021-02-19 17:19:36 +08006971static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data,
6972 struct io_ring_ctx *ctx,
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006973 void (*rsrc_put)(struct io_ring_ctx *ctx,
6974 struct io_rsrc_put *prsrc))
Hao Xu8bad28d2021-02-19 17:19:36 +08006975{
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006976 struct fixed_rsrc_ref_node *backup_node;
Hao Xu8bad28d2021-02-19 17:19:36 +08006977 int ret;
Xiaoguang Wang05589552020-03-31 14:05:18 +08006978
Hao Xu8bad28d2021-02-19 17:19:36 +08006979 if (data->quiesce)
6980 return -ENXIO;
6981
6982 data->quiesce = true;
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00006983 do {
Pavel Begunkovf2303b12021-02-20 18:03:49 +00006984 ret = -ENOMEM;
6985 backup_node = alloc_fixed_rsrc_ref_node(ctx);
6986 if (!backup_node)
6987 break;
6988 backup_node->rsrc_data = data;
6989 backup_node->rsrc_put = rsrc_put;
6990
Hao Xu8bad28d2021-02-19 17:19:36 +08006991 io_sqe_rsrc_kill_node(ctx, data);
6992 percpu_ref_kill(&data->refs);
6993 flush_delayed_work(&ctx->rsrc_put_work);
6994
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00006995 ret = wait_for_completion_interruptible(&data->done);
6996 if (!ret)
6997 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006998
Jens Axboecb5e1b82021-02-25 07:37:35 -07006999 percpu_ref_resurrect(&data->refs);
Hao Xu8bad28d2021-02-19 17:19:36 +08007000 io_sqe_rsrc_set_node(ctx, data, backup_node);
7001 backup_node = NULL;
Jens Axboecb5e1b82021-02-25 07:37:35 -07007002 reinit_completion(&data->done);
Hao Xu8bad28d2021-02-19 17:19:36 +08007003 mutex_unlock(&ctx->uring_lock);
7004 ret = io_run_task_work_sig();
7005 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007006 } while (ret >= 0);
Hao Xu8bad28d2021-02-19 17:19:36 +08007007 data->quiesce = false;
7008
7009 if (backup_node)
7010 destroy_fixed_rsrc_ref_node(backup_node);
7011 return ret;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007012}
7013
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007014static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx)
7015{
7016 struct fixed_rsrc_data *data;
7017
7018 data = kzalloc(sizeof(*data), GFP_KERNEL);
7019 if (!data)
7020 return NULL;
7021
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007022 if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero,
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007023 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) {
7024 kfree(data);
7025 return NULL;
7026 }
7027 data->ctx = ctx;
7028 init_completion(&data->done);
7029 return data;
7030}
7031
7032static void free_fixed_rsrc_data(struct fixed_rsrc_data *data)
7033{
7034 percpu_ref_exit(&data->refs);
7035 kfree(data->table);
7036 kfree(data);
7037}
7038
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007039static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
7040{
7041 struct fixed_rsrc_data *data = ctx->file_data;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007042 unsigned nr_tables, i;
7043 int ret;
7044
Hao Xu8bad28d2021-02-19 17:19:36 +08007045 /*
7046 * percpu_ref_is_dying() is to stop parallel files unregister
7047 * Since we possibly drop uring lock later in this function to
7048 * run task work.
7049 */
7050 if (!data || percpu_ref_is_dying(&data->refs))
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007051 return -ENXIO;
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007052 ret = io_rsrc_ref_quiesce(data, ctx, io_ring_file_put);
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007053 if (ret)
7054 return ret;
7055
Jens Axboe6b063142019-01-10 22:13:58 -07007056 __io_sqe_files_unregister(ctx);
Jens Axboe65e19f52019-10-26 07:20:21 -06007057 nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
7058 for (i = 0; i < nr_tables; i++)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007059 kfree(data->table[i].files);
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007060 free_fixed_rsrc_data(data);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007061 ctx->file_data = NULL;
Jens Axboe6b063142019-01-10 22:13:58 -07007062 ctx->nr_user_files = 0;
7063 return 0;
7064}
7065
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007066static void io_sq_thread_unpark(struct io_sq_data *sqd)
7067 __releases(&sqd->lock)
7068{
7069 if (!sqd->thread)
7070 return;
7071 if (sqd->thread == current)
7072 return;
7073 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
7074 wake_up_state(sqd->thread, TASK_PARKED);
7075 mutex_unlock(&sqd->lock);
7076}
7077
7078static bool io_sq_thread_park(struct io_sq_data *sqd)
7079 __acquires(&sqd->lock)
7080{
7081 if (sqd->thread == current)
7082 return true;
7083 mutex_lock(&sqd->lock);
7084 if (!sqd->thread) {
7085 mutex_unlock(&sqd->lock);
7086 return false;
7087 }
7088 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
7089 wake_up_process(sqd->thread);
7090 wait_for_completion(&sqd->completion);
7091 return true;
7092}
7093
7094static void io_sq_thread_stop(struct io_sq_data *sqd)
7095{
Jens Axboee54945a2021-02-26 11:27:15 -07007096 if (test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state))
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007097 return;
Jens Axboee54945a2021-02-26 11:27:15 -07007098 mutex_lock(&sqd->lock);
7099 if (sqd->thread) {
7100 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
7101 WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state));
7102 wake_up_process(sqd->thread);
7103 mutex_unlock(&sqd->lock);
7104 wait_for_completion(&sqd->exited);
7105 WARN_ON_ONCE(sqd->thread);
7106 } else {
7107 mutex_unlock(&sqd->lock);
7108 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007109}
7110
Jens Axboe534ca6d2020-09-02 13:52:19 -06007111static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007112{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007113 if (refcount_dec_and_test(&sqd->refs)) {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007114 io_sq_thread_stop(sqd);
7115 kfree(sqd);
7116 }
7117}
7118
7119static void io_sq_thread_finish(struct io_ring_ctx *ctx)
7120{
7121 struct io_sq_data *sqd = ctx->sq_data;
7122
7123 if (sqd) {
Jens Axboeeb858902021-02-25 10:13:29 -07007124 complete(&sqd->startup);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007125 if (sqd->thread) {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007126 wait_for_completion(&ctx->sq_thread_comp);
7127 io_sq_thread_park(sqd);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007128 }
7129
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007130 mutex_lock(&sqd->ctx_lock);
7131 list_del(&ctx->sqd_list);
7132 io_sqd_update_thread_idle(sqd);
7133 mutex_unlock(&sqd->ctx_lock);
7134
7135 if (sqd->thread)
7136 io_sq_thread_unpark(sqd);
7137
7138 io_put_sq_data(sqd);
7139 ctx->sq_data = NULL;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007140 }
7141}
7142
Jens Axboeaa061652020-09-02 14:50:27 -06007143static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7144{
7145 struct io_ring_ctx *ctx_attach;
7146 struct io_sq_data *sqd;
7147 struct fd f;
7148
7149 f = fdget(p->wq_fd);
7150 if (!f.file)
7151 return ERR_PTR(-ENXIO);
7152 if (f.file->f_op != &io_uring_fops) {
7153 fdput(f);
7154 return ERR_PTR(-EINVAL);
7155 }
7156
7157 ctx_attach = f.file->private_data;
7158 sqd = ctx_attach->sq_data;
7159 if (!sqd) {
7160 fdput(f);
7161 return ERR_PTR(-EINVAL);
7162 }
7163
7164 refcount_inc(&sqd->refs);
7165 fdput(f);
7166 return sqd;
7167}
7168
Jens Axboe534ca6d2020-09-02 13:52:19 -06007169static struct io_sq_data *io_get_sq_data(struct io_uring_params *p)
7170{
7171 struct io_sq_data *sqd;
7172
Jens Axboeaa061652020-09-02 14:50:27 -06007173 if (p->flags & IORING_SETUP_ATTACH_WQ)
7174 return io_attach_sq_data(p);
7175
Jens Axboe534ca6d2020-09-02 13:52:19 -06007176 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7177 if (!sqd)
7178 return ERR_PTR(-ENOMEM);
7179
7180 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007181 INIT_LIST_HEAD(&sqd->ctx_list);
7182 INIT_LIST_HEAD(&sqd->ctx_new_list);
7183 mutex_init(&sqd->ctx_lock);
7184 mutex_init(&sqd->lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007185 init_waitqueue_head(&sqd->wait);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007186 init_completion(&sqd->startup);
7187 init_completion(&sqd->completion);
7188 init_completion(&sqd->exited);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007189 return sqd;
7190}
7191
Jens Axboe6b063142019-01-10 22:13:58 -07007192#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007193/*
7194 * Ensure the UNIX gc is aware of our file set, so we are certain that
7195 * the io_uring can be safely unregistered on process exit, even if we have
7196 * loops in the file referencing.
7197 */
7198static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7199{
7200 struct sock *sk = ctx->ring_sock->sk;
7201 struct scm_fp_list *fpl;
7202 struct sk_buff *skb;
Jens Axboe08a45172019-10-03 08:11:03 -06007203 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007204
Jens Axboe6b063142019-01-10 22:13:58 -07007205 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7206 if (!fpl)
7207 return -ENOMEM;
7208
7209 skb = alloc_skb(0, GFP_KERNEL);
7210 if (!skb) {
7211 kfree(fpl);
7212 return -ENOMEM;
7213 }
7214
7215 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007216
Jens Axboe08a45172019-10-03 08:11:03 -06007217 nr_files = 0;
Jens Axboe62e398b2021-02-21 16:19:37 -07007218 fpl->user = get_uid(current_user());
Jens Axboe6b063142019-01-10 22:13:58 -07007219 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007220 struct file *file = io_file_from_index(ctx, i + offset);
7221
7222 if (!file)
Jens Axboe08a45172019-10-03 08:11:03 -06007223 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007224 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a45172019-10-03 08:11:03 -06007225 unix_inflight(fpl->user, fpl->fp[nr_files]);
7226 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007227 }
7228
Jens Axboe08a45172019-10-03 08:11:03 -06007229 if (nr_files) {
7230 fpl->max = SCM_MAX_FD;
7231 fpl->count = nr_files;
7232 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007233 skb->destructor = unix_destruct_scm;
Jens Axboe08a45172019-10-03 08:11:03 -06007234 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7235 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007236
Jens Axboe08a45172019-10-03 08:11:03 -06007237 for (i = 0; i < nr_files; i++)
7238 fput(fpl->fp[i]);
7239 } else {
7240 kfree_skb(skb);
7241 kfree(fpl);
7242 }
Jens Axboe6b063142019-01-10 22:13:58 -07007243
7244 return 0;
7245}
7246
7247/*
7248 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7249 * causes regular reference counting to break down. We rely on the UNIX
7250 * garbage collection to take care of this problem for us.
7251 */
7252static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7253{
7254 unsigned left, total;
7255 int ret = 0;
7256
7257 total = 0;
7258 left = ctx->nr_user_files;
7259 while (left) {
7260 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
Jens Axboe6b063142019-01-10 22:13:58 -07007261
7262 ret = __io_sqe_files_scm(ctx, this_files, total);
7263 if (ret)
7264 break;
7265 left -= this_files;
7266 total += this_files;
7267 }
7268
7269 if (!ret)
7270 return 0;
7271
7272 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007273 struct file *file = io_file_from_index(ctx, total);
7274
7275 if (file)
7276 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007277 total++;
7278 }
7279
7280 return ret;
7281}
7282#else
7283static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7284{
7285 return 0;
7286}
7287#endif
7288
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007289static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data,
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007290 unsigned nr_tables, unsigned nr_files)
Jens Axboe65e19f52019-10-26 07:20:21 -06007291{
7292 int i;
7293
7294 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007295 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007296 unsigned this_files;
7297
7298 this_files = min(nr_files, IORING_MAX_FILES_TABLE);
7299 table->files = kcalloc(this_files, sizeof(struct file *),
7300 GFP_KERNEL);
7301 if (!table->files)
7302 break;
7303 nr_files -= this_files;
7304 }
7305
7306 if (i == nr_tables)
7307 return 0;
7308
7309 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007310 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007311 kfree(table->files);
7312 }
7313 return 1;
7314}
7315
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007316static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
Jens Axboec3a31e62019-10-03 13:59:56 -06007317{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007318 struct file *file = prsrc->file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007319#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007320 struct sock *sock = ctx->ring_sock->sk;
7321 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7322 struct sk_buff *skb;
7323 int i;
7324
7325 __skb_queue_head_init(&list);
7326
7327 /*
7328 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7329 * remove this entry and rearrange the file array.
7330 */
7331 skb = skb_dequeue(head);
7332 while (skb) {
7333 struct scm_fp_list *fp;
7334
7335 fp = UNIXCB(skb).fp;
7336 for (i = 0; i < fp->count; i++) {
7337 int left;
7338
7339 if (fp->fp[i] != file)
7340 continue;
7341
7342 unix_notinflight(fp->user, fp->fp[i]);
7343 left = fp->count - 1 - i;
7344 if (left) {
7345 memmove(&fp->fp[i], &fp->fp[i + 1],
7346 left * sizeof(struct file *));
7347 }
7348 fp->count--;
7349 if (!fp->count) {
7350 kfree_skb(skb);
7351 skb = NULL;
7352 } else {
7353 __skb_queue_tail(&list, skb);
7354 }
7355 fput(file);
7356 file = NULL;
7357 break;
7358 }
7359
7360 if (!file)
7361 break;
7362
7363 __skb_queue_tail(&list, skb);
7364
7365 skb = skb_dequeue(head);
7366 }
7367
7368 if (skb_peek(&list)) {
7369 spin_lock_irq(&head->lock);
7370 while ((skb = __skb_dequeue(&list)) != NULL)
7371 __skb_queue_tail(head, skb);
7372 spin_unlock_irq(&head->lock);
7373 }
7374#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007375 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007376#endif
7377}
7378
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007379static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007380{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007381 struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data;
7382 struct io_ring_ctx *ctx = rsrc_data->ctx;
7383 struct io_rsrc_put *prsrc, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007384
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007385 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) {
7386 list_del(&prsrc->list);
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007387 ref_node->rsrc_put(ctx, prsrc);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007388 kfree(prsrc);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007389 }
7390
Xiaoguang Wang05589552020-03-31 14:05:18 +08007391 percpu_ref_exit(&ref_node->refs);
7392 kfree(ref_node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007393 percpu_ref_put(&rsrc_data->refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007394}
7395
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007396static void io_rsrc_put_work(struct work_struct *work)
Jens Axboe4a38aed22020-05-14 17:21:15 -06007397{
7398 struct io_ring_ctx *ctx;
7399 struct llist_node *node;
7400
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007401 ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work);
7402 node = llist_del_all(&ctx->rsrc_put_llist);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007403
7404 while (node) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007405 struct fixed_rsrc_ref_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007406 struct llist_node *next = node->next;
7407
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007408 ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist);
7409 __io_rsrc_put_work(ref_node);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007410 node = next;
7411 }
7412}
7413
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007414static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data,
7415 unsigned i)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007416{
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007417 struct fixed_rsrc_table *table;
7418
7419 table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT];
7420 return &table->files[i & IORING_FILE_TABLE_MASK];
7421}
7422
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007423static void io_rsrc_node_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007424{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007425 struct fixed_rsrc_ref_node *ref_node;
7426 struct fixed_rsrc_data *data;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007427 struct io_ring_ctx *ctx;
Pavel Begunkove2978222020-11-18 14:56:26 +00007428 bool first_add = false;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007429 int delay = HZ;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007430
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007431 ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs);
7432 data = ref_node->rsrc_data;
Pavel Begunkove2978222020-11-18 14:56:26 +00007433 ctx = data->ctx;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007434
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007435 io_rsrc_ref_lock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007436 ref_node->done = true;
7437
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007438 while (!list_empty(&ctx->rsrc_ref_list)) {
7439 ref_node = list_first_entry(&ctx->rsrc_ref_list,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007440 struct fixed_rsrc_ref_node, node);
Pavel Begunkove2978222020-11-18 14:56:26 +00007441 /* recycle ref nodes in order */
7442 if (!ref_node->done)
7443 break;
7444 list_del(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007445 first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist);
Pavel Begunkove2978222020-11-18 14:56:26 +00007446 }
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007447 io_rsrc_ref_unlock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007448
7449 if (percpu_ref_is_dying(&data->refs))
Jens Axboe4a38aed22020-05-14 17:21:15 -06007450 delay = 0;
7451
Jens Axboe4a38aed22020-05-14 17:21:15 -06007452 if (!delay)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007453 mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007454 else if (first_add)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007455 queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007456}
7457
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007458static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Xiaoguang Wang05589552020-03-31 14:05:18 +08007459 struct io_ring_ctx *ctx)
7460{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007461 struct fixed_rsrc_ref_node *ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007462
7463 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7464 if (!ref_node)
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007465 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007466
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007467 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007468 0, GFP_KERNEL)) {
7469 kfree(ref_node);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007470 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007471 }
7472 INIT_LIST_HEAD(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007473 INIT_LIST_HEAD(&ref_node->rsrc_list);
Pavel Begunkove2978222020-11-18 14:56:26 +00007474 ref_node->done = false;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007475 return ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007476}
7477
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007478static void init_fixed_file_ref_node(struct io_ring_ctx *ctx,
7479 struct fixed_rsrc_ref_node *ref_node)
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007480{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007481 ref_node->rsrc_data = ctx->file_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007482 ref_node->rsrc_put = io_ring_file_put;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007483}
7484
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007485static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node)
Xiaoguang Wang05589552020-03-31 14:05:18 +08007486{
7487 percpu_ref_exit(&ref_node->refs);
7488 kfree(ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007489}
7490
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007491
Jens Axboe05f3fb32019-12-09 11:22:50 -07007492static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
7493 unsigned nr_args)
7494{
7495 __s32 __user *fds = (__s32 __user *) arg;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007496 unsigned nr_tables, i;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007497 struct file *file;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007498 int fd, ret = -ENOMEM;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007499 struct fixed_rsrc_ref_node *ref_node;
7500 struct fixed_rsrc_data *file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007501
7502 if (ctx->file_data)
7503 return -EBUSY;
7504 if (!nr_args)
7505 return -EINVAL;
7506 if (nr_args > IORING_MAX_FIXED_FILES)
7507 return -EMFILE;
7508
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007509 file_data = alloc_fixed_rsrc_data(ctx);
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007510 if (!file_data)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007511 return -ENOMEM;
Dan Carpenter13770a72021-02-01 15:23:42 +03007512 ctx->file_data = file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007513
7514 nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE);
Colin Ian King035fbaf2020-10-12 15:03:41 +01007515 file_data->table = kcalloc(nr_tables, sizeof(*file_data->table),
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007516 GFP_KERNEL);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007517 if (!file_data->table)
7518 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007519
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007520 if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args))
Jens Axboe05f3fb32019-12-09 11:22:50 -07007521 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007522
Jens Axboe05f3fb32019-12-09 11:22:50 -07007523 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007524 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
7525 ret = -EFAULT;
7526 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007527 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007528 /* allow sparse sets */
7529 if (fd == -1)
7530 continue;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007531
Jens Axboe05f3fb32019-12-09 11:22:50 -07007532 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007533 ret = -EBADF;
7534 if (!file)
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007535 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007536
7537 /*
7538 * Don't allow io_uring instances to be registered. If UNIX
7539 * isn't enabled, then this causes a reference cycle and this
7540 * instance can never get freed. If UNIX is enabled we'll
7541 * handle it just fine, but there's still no point in allowing
7542 * a ring fd as it doesn't support regular read/write anyway.
7543 */
7544 if (file->f_op == &io_uring_fops) {
7545 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007546 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007547 }
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007548 *io_fixed_file_slot(file_data, i) = file;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007549 }
7550
Jens Axboe05f3fb32019-12-09 11:22:50 -07007551 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007552 if (ret) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07007553 io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007554 return ret;
7555 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007556
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007557 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007558 if (!ref_node) {
Xiaoguang Wang05589552020-03-31 14:05:18 +08007559 io_sqe_files_unregister(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007560 return -ENOMEM;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007561 }
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007562 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007563
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007564 io_sqe_rsrc_set_node(ctx, file_data, ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007565 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007566out_fput:
7567 for (i = 0; i < ctx->nr_user_files; i++) {
7568 file = io_file_from_index(ctx, i);
7569 if (file)
7570 fput(file);
7571 }
7572 for (i = 0; i < nr_tables; i++)
7573 kfree(file_data->table[i].files);
7574 ctx->nr_user_files = 0;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007575out_free:
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007576 free_fixed_rsrc_data(ctx->file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007577 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007578 return ret;
7579}
7580
Jens Axboec3a31e62019-10-03 13:59:56 -06007581static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7582 int index)
7583{
7584#if defined(CONFIG_UNIX)
7585 struct sock *sock = ctx->ring_sock->sk;
7586 struct sk_buff_head *head = &sock->sk_receive_queue;
7587 struct sk_buff *skb;
7588
7589 /*
7590 * See if we can merge this file into an existing skb SCM_RIGHTS
7591 * file set. If there's no room, fall back to allocating a new skb
7592 * and filling it in.
7593 */
7594 spin_lock_irq(&head->lock);
7595 skb = skb_peek(head);
7596 if (skb) {
7597 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7598
7599 if (fpl->count < SCM_MAX_FD) {
7600 __skb_unlink(skb, head);
7601 spin_unlock_irq(&head->lock);
7602 fpl->fp[fpl->count] = get_file(file);
7603 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7604 fpl->count++;
7605 spin_lock_irq(&head->lock);
7606 __skb_queue_head(head, skb);
7607 } else {
7608 skb = NULL;
7609 }
7610 }
7611 spin_unlock_irq(&head->lock);
7612
7613 if (skb) {
7614 fput(file);
7615 return 0;
7616 }
7617
7618 return __io_sqe_files_scm(ctx, 1, index);
7619#else
7620 return 0;
7621#endif
7622}
7623
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007624static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007625{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007626 struct io_rsrc_put *prsrc;
7627 struct fixed_rsrc_ref_node *ref_node = data->node;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007628
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007629 prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
7630 if (!prsrc)
Hillf Dantona5318d32020-03-23 17:47:15 +08007631 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007632
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007633 prsrc->rsrc = rsrc;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007634 list_add(&prsrc->list, &ref_node->rsrc_list);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007635
Hillf Dantona5318d32020-03-23 17:47:15 +08007636 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007637}
7638
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007639static inline int io_queue_file_removal(struct fixed_rsrc_data *data,
7640 struct file *file)
7641{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007642 return io_queue_rsrc_removal(data, (void *)file);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007643}
7644
Jens Axboe05f3fb32019-12-09 11:22:50 -07007645static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007646 struct io_uring_rsrc_update *up,
Jens Axboe05f3fb32019-12-09 11:22:50 -07007647 unsigned nr_args)
7648{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007649 struct fixed_rsrc_data *data = ctx->file_data;
7650 struct fixed_rsrc_ref_node *ref_node;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007651 struct file *file, **file_slot;
Jens Axboec3a31e62019-10-03 13:59:56 -06007652 __s32 __user *fds;
7653 int fd, i, err;
7654 __u32 done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007655 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06007656
Jens Axboe05f3fb32019-12-09 11:22:50 -07007657 if (check_add_overflow(up->offset, nr_args, &done))
Jens Axboec3a31e62019-10-03 13:59:56 -06007658 return -EOVERFLOW;
7659 if (done > ctx->nr_user_files)
7660 return -EINVAL;
7661
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007662 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007663 if (!ref_node)
7664 return -ENOMEM;
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007665 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007666
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007667 fds = u64_to_user_ptr(up->data);
Pavel Begunkov67973b92021-01-26 13:51:09 +00007668 for (done = 0; done < nr_args; done++) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007669 err = 0;
7670 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
7671 err = -EFAULT;
7672 break;
7673 }
noah4e0377a2021-01-26 15:23:28 -05007674 if (fd == IORING_REGISTER_FILES_SKIP)
7675 continue;
7676
Pavel Begunkov67973b92021-01-26 13:51:09 +00007677 i = array_index_nospec(up->offset + done, ctx->nr_user_files);
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007678 file_slot = io_fixed_file_slot(ctx->file_data, i);
7679
7680 if (*file_slot) {
7681 err = io_queue_file_removal(data, *file_slot);
Hillf Dantona5318d32020-03-23 17:47:15 +08007682 if (err)
7683 break;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007684 *file_slot = NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007685 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06007686 }
7687 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007688 file = fget(fd);
7689 if (!file) {
7690 err = -EBADF;
7691 break;
7692 }
7693 /*
7694 * Don't allow io_uring instances to be registered. If
7695 * UNIX isn't enabled, then this causes a reference
7696 * cycle and this instance can never get freed. If UNIX
7697 * is enabled we'll handle it just fine, but there's
7698 * still no point in allowing a ring fd as it doesn't
7699 * support regular read/write anyway.
7700 */
7701 if (file->f_op == &io_uring_fops) {
7702 fput(file);
7703 err = -EBADF;
7704 break;
7705 }
Jens Axboee68a3ff2021-02-11 07:45:08 -07007706 *file_slot = file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007707 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007708 if (err) {
Jens Axboee68a3ff2021-02-11 07:45:08 -07007709 *file_slot = NULL;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007710 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007711 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007712 }
Jens Axboec3a31e62019-10-03 13:59:56 -06007713 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007714 }
7715
Xiaoguang Wang05589552020-03-31 14:05:18 +08007716 if (needs_switch) {
Pavel Begunkovb2e96852020-10-10 18:34:16 +01007717 percpu_ref_kill(&data->node->refs);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007718 io_sqe_rsrc_set_node(ctx, data, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007719 } else
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007720 destroy_fixed_rsrc_ref_node(ref_node);
Jens Axboec3a31e62019-10-03 13:59:56 -06007721
7722 return done ? done : err;
7723}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007724
Jens Axboe05f3fb32019-12-09 11:22:50 -07007725static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
7726 unsigned nr_args)
7727{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007728 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007729
7730 if (!ctx->file_data)
7731 return -ENXIO;
7732 if (!nr_args)
7733 return -EINVAL;
7734 if (copy_from_user(&up, arg, sizeof(up)))
7735 return -EFAULT;
7736 if (up.resv)
7737 return -EINVAL;
7738
7739 return __io_sqe_files_update(ctx, &up, nr_args);
7740}
Jens Axboec3a31e62019-10-03 13:59:56 -06007741
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007742static struct io_wq_work *io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07007743{
7744 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
7745
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007746 req = io_put_req_find_next(req);
7747 return req ? &req->work : NULL;
Jens Axboe7d723062019-11-12 22:31:31 -07007748}
7749
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007750static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx)
Pavel Begunkov24369c22020-01-28 03:15:48 +03007751{
Jens Axboee9418942021-02-19 12:33:30 -07007752 struct io_wq_hash *hash;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007753 struct io_wq_data data;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007754 unsigned int concurrency;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007755
Jens Axboee9418942021-02-19 12:33:30 -07007756 hash = ctx->hash_map;
7757 if (!hash) {
7758 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
7759 if (!hash)
7760 return ERR_PTR(-ENOMEM);
7761 refcount_set(&hash->refs, 1);
7762 init_waitqueue_head(&hash->wait);
7763 ctx->hash_map = hash;
7764 }
7765
7766 data.hash = hash;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007767 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03007768 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007769
Jens Axboed25e3a32021-02-16 11:41:41 -07007770 /* Do QD, or 4 * CPUS, whatever is smallest */
7771 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
Pavel Begunkov24369c22020-01-28 03:15:48 +03007772
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007773 return io_wq_create(concurrency, &data);
Pavel Begunkov24369c22020-01-28 03:15:48 +03007774}
7775
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007776static int io_uring_alloc_task_context(struct task_struct *task,
7777 struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06007778{
7779 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06007780 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06007781
7782 tctx = kmalloc(sizeof(*tctx), GFP_KERNEL);
7783 if (unlikely(!tctx))
7784 return -ENOMEM;
7785
Jens Axboed8a6df12020-10-15 16:24:45 -06007786 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
7787 if (unlikely(ret)) {
7788 kfree(tctx);
7789 return ret;
7790 }
7791
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007792 tctx->io_wq = io_init_wq_offload(ctx);
7793 if (IS_ERR(tctx->io_wq)) {
7794 ret = PTR_ERR(tctx->io_wq);
7795 percpu_counter_destroy(&tctx->inflight);
7796 kfree(tctx);
7797 return ret;
7798 }
7799
Jens Axboe0f212202020-09-13 13:09:39 -06007800 xa_init(&tctx->xa);
7801 init_waitqueue_head(&tctx->wait);
7802 tctx->last = NULL;
Jens Axboefdaf0832020-10-30 09:37:30 -06007803 atomic_set(&tctx->in_idle, 0);
7804 tctx->sqpoll = false;
Jens Axboe0f212202020-09-13 13:09:39 -06007805 task->io_uring = tctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00007806 spin_lock_init(&tctx->task_lock);
7807 INIT_WQ_LIST(&tctx->task_list);
7808 tctx->task_state = 0;
7809 init_task_work(&tctx->task_work, tctx_task_work);
Jens Axboe0f212202020-09-13 13:09:39 -06007810 return 0;
7811}
7812
7813void __io_uring_free(struct task_struct *tsk)
7814{
7815 struct io_uring_task *tctx = tsk->io_uring;
7816
7817 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Pavel Begunkovef8eaa42021-02-27 11:16:45 +00007818 WARN_ON_ONCE(tctx->io_wq);
7819
Jens Axboed8a6df12020-10-15 16:24:45 -06007820 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06007821 kfree(tctx);
7822 tsk->io_uring = NULL;
7823}
7824
Jens Axboe5f3f26f2021-02-25 10:17:46 -07007825static int io_sq_thread_fork(struct io_sq_data *sqd, struct io_ring_ctx *ctx)
7826{
Jens Axboe46fe18b2021-03-04 12:39:36 -07007827 struct task_struct *tsk;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07007828 int ret;
7829
7830 clear_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
7831 reinit_completion(&sqd->completion);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00007832 ctx->sqo_exec = 0;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07007833 sqd->task_pid = current->pid;
Jens Axboe46fe18b2021-03-04 12:39:36 -07007834 tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
7835 if (IS_ERR(tsk))
7836 return PTR_ERR(tsk);
7837 ret = io_uring_alloc_task_context(tsk, ctx);
7838 if (ret)
7839 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
7840 sqd->thread = tsk;
7841 wake_up_new_task(tsk);
7842 return ret;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07007843}
7844
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007845static int io_sq_offload_create(struct io_ring_ctx *ctx,
7846 struct io_uring_params *p)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007847{
7848 int ret;
7849
Jens Axboed25e3a32021-02-16 11:41:41 -07007850 /* Retain compatibility with failing for an invalid attach attempt */
7851 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
7852 IORING_SETUP_ATTACH_WQ) {
7853 struct fd f;
7854
7855 f = fdget(p->wq_fd);
7856 if (!f.file)
7857 return -ENXIO;
7858 if (f.file->f_op != &io_uring_fops) {
7859 fdput(f);
7860 return -EINVAL;
7861 }
7862 fdput(f);
7863 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07007864 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe46fe18b2021-03-04 12:39:36 -07007865 struct task_struct *tsk;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007866 struct io_sq_data *sqd;
7867
Jens Axboe3ec482d2019-04-08 10:51:01 -06007868 ret = -EPERM;
Jens Axboece59fc62020-09-02 13:28:09 -06007869 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE))
Jens Axboe3ec482d2019-04-08 10:51:01 -06007870 goto err;
7871
Jens Axboe534ca6d2020-09-02 13:52:19 -06007872 sqd = io_get_sq_data(p);
7873 if (IS_ERR(sqd)) {
7874 ret = PTR_ERR(sqd);
7875 goto err;
7876 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007877
Jens Axboe534ca6d2020-09-02 13:52:19 -06007878 ctx->sq_data = sqd;
Jens Axboe69fb2132020-09-14 11:16:23 -06007879 io_sq_thread_park(sqd);
7880 mutex_lock(&sqd->ctx_lock);
7881 list_add(&ctx->sqd_list, &sqd->ctx_new_list);
7882 mutex_unlock(&sqd->ctx_lock);
7883 io_sq_thread_unpark(sqd);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007884
Jens Axboe917257d2019-04-13 09:28:55 -06007885 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
7886 if (!ctx->sq_thread_idle)
7887 ctx->sq_thread_idle = HZ;
7888
Jens Axboeaa061652020-09-02 14:50:27 -06007889 if (sqd->thread)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007890 return 0;
Jens Axboeaa061652020-09-02 14:50:27 -06007891
Jens Axboe6c271ce2019-01-10 11:22:30 -07007892 if (p->flags & IORING_SETUP_SQ_AFF) {
Jens Axboe44a9bd12019-05-14 20:00:30 -06007893 int cpu = p->sq_thread_cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007894
Jens Axboe917257d2019-04-13 09:28:55 -06007895 ret = -EINVAL;
Jens Axboe44a9bd12019-05-14 20:00:30 -06007896 if (cpu >= nr_cpu_ids)
7897 goto err;
Shenghui Wang7889f442019-05-07 16:03:19 +08007898 if (!cpu_online(cpu))
Jens Axboe917257d2019-04-13 09:28:55 -06007899 goto err;
7900
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007901 sqd->sq_cpu = cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007902 } else {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007903 sqd->sq_cpu = -1;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007904 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007905
7906 sqd->task_pid = current->pid;
Jens Axboe46fe18b2021-03-04 12:39:36 -07007907 tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
7908 if (IS_ERR(tsk)) {
7909 ret = PTR_ERR(tsk);
Jens Axboe6c271ce2019-01-10 11:22:30 -07007910 goto err;
7911 }
Jens Axboe46fe18b2021-03-04 12:39:36 -07007912 ret = io_uring_alloc_task_context(tsk, ctx);
7913 if (ret)
7914 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
7915 sqd->thread = tsk;
7916 wake_up_new_task(tsk);
Jens Axboe0f212202020-09-13 13:09:39 -06007917 if (ret)
7918 goto err;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007919 } else if (p->flags & IORING_SETUP_SQ_AFF) {
7920 /* Can't have SQ_AFF without SQPOLL */
7921 ret = -EINVAL;
7922 goto err;
7923 }
7924
Jens Axboe2b188cc2019-01-07 10:46:33 -07007925 return 0;
7926err:
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007927 io_sq_thread_finish(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007928 return ret;
7929}
7930
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007931static void io_sq_offload_start(struct io_ring_ctx *ctx)
7932{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007933 struct io_sq_data *sqd = ctx->sq_data;
7934
Jens Axboe3ebba792021-02-28 15:32:18 -07007935 ctx->flags &= ~IORING_SETUP_R_DISABLED;
Jens Axboeeb858902021-02-25 10:13:29 -07007936 if (ctx->flags & IORING_SETUP_SQPOLL)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007937 complete(&sqd->startup);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007938}
7939
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007940static inline void __io_unaccount_mem(struct user_struct *user,
7941 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007942{
7943 atomic_long_sub(nr_pages, &user->locked_vm);
7944}
7945
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007946static inline int __io_account_mem(struct user_struct *user,
7947 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007948{
7949 unsigned long page_limit, cur_pages, new_pages;
7950
7951 /* Don't allow more pages than we can safely lock */
7952 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
7953
7954 do {
7955 cur_pages = atomic_long_read(&user->locked_vm);
7956 new_pages = cur_pages + nr_pages;
7957 if (new_pages > page_limit)
7958 return -ENOMEM;
7959 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
7960 new_pages) != cur_pages);
7961
7962 return 0;
7963}
7964
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007965static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007966{
Jens Axboe62e398b2021-02-21 16:19:37 -07007967 if (ctx->user)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007968 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007969
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007970 if (ctx->mm_account)
7971 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007972}
7973
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007974static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007975{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007976 int ret;
7977
Jens Axboe62e398b2021-02-21 16:19:37 -07007978 if (ctx->user) {
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007979 ret = __io_account_mem(ctx->user, nr_pages);
7980 if (ret)
7981 return ret;
7982 }
7983
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007984 if (ctx->mm_account)
7985 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007986
7987 return 0;
7988}
7989
Jens Axboe2b188cc2019-01-07 10:46:33 -07007990static void io_mem_free(void *ptr)
7991{
Mark Rutland52e04ef2019-04-30 17:30:21 +01007992 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007993
Mark Rutland52e04ef2019-04-30 17:30:21 +01007994 if (!ptr)
7995 return;
7996
7997 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007998 if (put_page_testzero(page))
7999 free_compound_page(page);
8000}
8001
8002static void *io_mem_alloc(size_t size)
8003{
8004 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008005 __GFP_NORETRY | __GFP_ACCOUNT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008006
8007 return (void *) __get_free_pages(gfp_flags, get_order(size));
8008}
8009
Hristo Venev75b28af2019-08-26 17:23:46 +00008010static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
8011 size_t *sq_offset)
8012{
8013 struct io_rings *rings;
8014 size_t off, sq_array_size;
8015
8016 off = struct_size(rings, cqes, cq_entries);
8017 if (off == SIZE_MAX)
8018 return SIZE_MAX;
8019
8020#ifdef CONFIG_SMP
8021 off = ALIGN(off, SMP_CACHE_BYTES);
8022 if (off == 0)
8023 return SIZE_MAX;
8024#endif
8025
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02008026 if (sq_offset)
8027 *sq_offset = off;
8028
Hristo Venev75b28af2019-08-26 17:23:46 +00008029 sq_array_size = array_size(sizeof(u32), sq_entries);
8030 if (sq_array_size == SIZE_MAX)
8031 return SIZE_MAX;
8032
8033 if (check_add_overflow(off, sq_array_size, &off))
8034 return SIZE_MAX;
8035
Hristo Venev75b28af2019-08-26 17:23:46 +00008036 return off;
8037}
8038
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008039static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
Jens Axboeedafcce2019-01-09 09:16:05 -07008040{
8041 int i, j;
8042
8043 if (!ctx->user_bufs)
8044 return -ENXIO;
8045
8046 for (i = 0; i < ctx->nr_user_bufs; i++) {
8047 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8048
8049 for (j = 0; j < imu->nr_bvecs; j++)
John Hubbardf1f6a7d2020-01-30 22:13:35 -08008050 unpin_user_page(imu->bvec[j].bv_page);
Jens Axboeedafcce2019-01-09 09:16:05 -07008051
Jens Axboede293932020-09-17 16:19:16 -06008052 if (imu->acct_pages)
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008053 io_unaccount_mem(ctx, imu->acct_pages);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008054 kvfree(imu->bvec);
Jens Axboeedafcce2019-01-09 09:16:05 -07008055 imu->nr_bvecs = 0;
8056 }
8057
8058 kfree(ctx->user_bufs);
8059 ctx->user_bufs = NULL;
8060 ctx->nr_user_bufs = 0;
8061 return 0;
8062}
8063
8064static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
8065 void __user *arg, unsigned index)
8066{
8067 struct iovec __user *src;
8068
8069#ifdef CONFIG_COMPAT
8070 if (ctx->compat) {
8071 struct compat_iovec __user *ciovs;
8072 struct compat_iovec ciov;
8073
8074 ciovs = (struct compat_iovec __user *) arg;
8075 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
8076 return -EFAULT;
8077
Jens Axboed55e5f52019-12-11 16:12:15 -07008078 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07008079 dst->iov_len = ciov.iov_len;
8080 return 0;
8081 }
8082#endif
8083 src = (struct iovec __user *) arg;
8084 if (copy_from_user(dst, &src[index], sizeof(*dst)))
8085 return -EFAULT;
8086 return 0;
8087}
8088
Jens Axboede293932020-09-17 16:19:16 -06008089/*
8090 * Not super efficient, but this is just a registration time. And we do cache
8091 * the last compound head, so generally we'll only do a full search if we don't
8092 * match that one.
8093 *
8094 * We check if the given compound head page has already been accounted, to
8095 * avoid double accounting it. This allows us to account the full size of the
8096 * page, not just the constituent pages of a huge page.
8097 */
8098static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8099 int nr_pages, struct page *hpage)
8100{
8101 int i, j;
8102
8103 /* check current page array */
8104 for (i = 0; i < nr_pages; i++) {
8105 if (!PageCompound(pages[i]))
8106 continue;
8107 if (compound_head(pages[i]) == hpage)
8108 return true;
8109 }
8110
8111 /* check previously registered pages */
8112 for (i = 0; i < ctx->nr_user_bufs; i++) {
8113 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8114
8115 for (j = 0; j < imu->nr_bvecs; j++) {
8116 if (!PageCompound(imu->bvec[j].bv_page))
8117 continue;
8118 if (compound_head(imu->bvec[j].bv_page) == hpage)
8119 return true;
8120 }
8121 }
8122
8123 return false;
8124}
8125
8126static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8127 int nr_pages, struct io_mapped_ubuf *imu,
8128 struct page **last_hpage)
8129{
8130 int i, ret;
8131
8132 for (i = 0; i < nr_pages; i++) {
8133 if (!PageCompound(pages[i])) {
8134 imu->acct_pages++;
8135 } else {
8136 struct page *hpage;
8137
8138 hpage = compound_head(pages[i]);
8139 if (hpage == *last_hpage)
8140 continue;
8141 *last_hpage = hpage;
8142 if (headpage_already_acct(ctx, pages, i, hpage))
8143 continue;
8144 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8145 }
8146 }
8147
8148 if (!imu->acct_pages)
8149 return 0;
8150
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008151 ret = io_account_mem(ctx, imu->acct_pages);
Jens Axboede293932020-09-17 16:19:16 -06008152 if (ret)
8153 imu->acct_pages = 0;
8154 return ret;
8155}
8156
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008157static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
8158 struct io_mapped_ubuf *imu,
8159 struct page **last_hpage)
Jens Axboeedafcce2019-01-09 09:16:05 -07008160{
8161 struct vm_area_struct **vmas = NULL;
8162 struct page **pages = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008163 unsigned long off, start, end, ubuf;
8164 size_t size;
8165 int ret, pret, nr_pages, i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008166
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008167 ubuf = (unsigned long) iov->iov_base;
8168 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8169 start = ubuf >> PAGE_SHIFT;
8170 nr_pages = end - start;
8171
8172 ret = -ENOMEM;
8173
8174 pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
8175 if (!pages)
8176 goto done;
8177
8178 vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *),
8179 GFP_KERNEL);
8180 if (!vmas)
8181 goto done;
8182
8183 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
8184 GFP_KERNEL);
8185 if (!imu->bvec)
8186 goto done;
8187
8188 ret = 0;
8189 mmap_read_lock(current->mm);
8190 pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
8191 pages, vmas);
8192 if (pret == nr_pages) {
8193 /* don't support file backed memory */
8194 for (i = 0; i < nr_pages; i++) {
8195 struct vm_area_struct *vma = vmas[i];
8196
8197 if (vma->vm_file &&
8198 !is_file_hugepages(vma->vm_file)) {
8199 ret = -EOPNOTSUPP;
8200 break;
8201 }
8202 }
8203 } else {
8204 ret = pret < 0 ? pret : -EFAULT;
8205 }
8206 mmap_read_unlock(current->mm);
8207 if (ret) {
8208 /*
8209 * if we did partial map, or found file backed vmas,
8210 * release any pages we did get
8211 */
8212 if (pret > 0)
8213 unpin_user_pages(pages, pret);
8214 kvfree(imu->bvec);
8215 goto done;
8216 }
8217
8218 ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage);
8219 if (ret) {
8220 unpin_user_pages(pages, pret);
8221 kvfree(imu->bvec);
8222 goto done;
8223 }
8224
8225 off = ubuf & ~PAGE_MASK;
8226 size = iov->iov_len;
8227 for (i = 0; i < nr_pages; i++) {
8228 size_t vec_len;
8229
8230 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8231 imu->bvec[i].bv_page = pages[i];
8232 imu->bvec[i].bv_len = vec_len;
8233 imu->bvec[i].bv_offset = off;
8234 off = 0;
8235 size -= vec_len;
8236 }
8237 /* store original address for later verification */
8238 imu->ubuf = ubuf;
8239 imu->len = iov->iov_len;
8240 imu->nr_bvecs = nr_pages;
8241 ret = 0;
8242done:
8243 kvfree(pages);
8244 kvfree(vmas);
8245 return ret;
8246}
8247
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008248static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008249{
Jens Axboeedafcce2019-01-09 09:16:05 -07008250 if (ctx->user_bufs)
8251 return -EBUSY;
8252 if (!nr_args || nr_args > UIO_MAXIOV)
8253 return -EINVAL;
8254
8255 ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
8256 GFP_KERNEL);
8257 if (!ctx->user_bufs)
8258 return -ENOMEM;
8259
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008260 return 0;
8261}
8262
8263static int io_buffer_validate(struct iovec *iov)
8264{
8265 /*
8266 * Don't impose further limits on the size and buffer
8267 * constraints here, we'll -EINVAL later when IO is
8268 * submitted if they are wrong.
8269 */
8270 if (!iov->iov_base || !iov->iov_len)
8271 return -EFAULT;
8272
8273 /* arbitrary limit, but we need something */
8274 if (iov->iov_len > SZ_1G)
8275 return -EFAULT;
8276
8277 return 0;
8278}
8279
8280static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
8281 unsigned int nr_args)
8282{
8283 int i, ret;
8284 struct iovec iov;
8285 struct page *last_hpage = NULL;
8286
8287 ret = io_buffers_map_alloc(ctx, nr_args);
8288 if (ret)
8289 return ret;
8290
Jens Axboeedafcce2019-01-09 09:16:05 -07008291 for (i = 0; i < nr_args; i++) {
8292 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
Jens Axboeedafcce2019-01-09 09:16:05 -07008293
8294 ret = io_copy_iov(ctx, &iov, arg, i);
8295 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008296 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008297
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008298 ret = io_buffer_validate(&iov);
8299 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008300 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008301
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008302 ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage);
8303 if (ret)
8304 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008305
8306 ctx->nr_user_bufs++;
8307 }
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008308
8309 if (ret)
8310 io_sqe_buffers_unregister(ctx);
8311
Jens Axboeedafcce2019-01-09 09:16:05 -07008312 return ret;
8313}
8314
Jens Axboe9b402842019-04-11 11:45:41 -06008315static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8316{
8317 __s32 __user *fds = arg;
8318 int fd;
8319
8320 if (ctx->cq_ev_fd)
8321 return -EBUSY;
8322
8323 if (copy_from_user(&fd, fds, sizeof(*fds)))
8324 return -EFAULT;
8325
8326 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8327 if (IS_ERR(ctx->cq_ev_fd)) {
8328 int ret = PTR_ERR(ctx->cq_ev_fd);
8329 ctx->cq_ev_fd = NULL;
8330 return ret;
8331 }
8332
8333 return 0;
8334}
8335
8336static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8337{
8338 if (ctx->cq_ev_fd) {
8339 eventfd_ctx_put(ctx->cq_ev_fd);
8340 ctx->cq_ev_fd = NULL;
8341 return 0;
8342 }
8343
8344 return -ENXIO;
8345}
8346
Jens Axboe5a2e7452020-02-23 16:23:11 -07008347static int __io_destroy_buffers(int id, void *p, void *data)
8348{
8349 struct io_ring_ctx *ctx = data;
8350 struct io_buffer *buf = p;
8351
Jens Axboe067524e2020-03-02 16:32:28 -07008352 __io_remove_buffers(ctx, buf, id, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008353 return 0;
8354}
8355
8356static void io_destroy_buffers(struct io_ring_ctx *ctx)
8357{
8358 idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx);
8359 idr_destroy(&ctx->io_buffer_idr);
8360}
8361
Jens Axboe68e68ee2021-02-13 09:00:02 -07008362static void io_req_cache_free(struct list_head *list, struct task_struct *tsk)
Jens Axboe1b4c3512021-02-10 00:03:19 +00008363{
Jens Axboe68e68ee2021-02-13 09:00:02 -07008364 struct io_kiocb *req, *nxt;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008365
Jens Axboe68e68ee2021-02-13 09:00:02 -07008366 list_for_each_entry_safe(req, nxt, list, compl.list) {
8367 if (tsk && req->task != tsk)
8368 continue;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008369 list_del(&req->compl.list);
8370 kmem_cache_free(req_cachep, req);
8371 }
8372}
8373
Jens Axboe4010fec2021-02-27 15:04:18 -07008374static void io_req_caches_free(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008375{
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008376 struct io_submit_state *submit_state = &ctx->submit_state;
Pavel Begunkove5547d22021-02-23 22:17:20 +00008377 struct io_comp_state *cs = &ctx->submit_state.comp;
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008378
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008379 mutex_lock(&ctx->uring_lock);
8380
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008381 if (submit_state->free_reqs) {
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008382 kmem_cache_free_bulk(req_cachep, submit_state->free_reqs,
8383 submit_state->reqs);
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008384 submit_state->free_reqs = 0;
8385 }
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008386
8387 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkove5547d22021-02-23 22:17:20 +00008388 list_splice_init(&cs->locked_free_list, &cs->free_list);
8389 cs->locked_free_nr = 0;
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008390 spin_unlock_irq(&ctx->completion_lock);
8391
Pavel Begunkove5547d22021-02-23 22:17:20 +00008392 io_req_cache_free(&cs->free_list, NULL);
8393
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008394 mutex_unlock(&ctx->uring_lock);
8395}
8396
Jens Axboe2b188cc2019-01-07 10:46:33 -07008397static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8398{
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00008399 /*
8400 * Some may use context even when all refs and requests have been put,
8401 * and they are free to do so while still holding uring_lock, see
8402 * __io_req_task_submit(). Wait for them to finish.
8403 */
8404 mutex_lock(&ctx->uring_lock);
8405 mutex_unlock(&ctx->uring_lock);
8406
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008407 io_sq_thread_finish(ctx);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008408 io_sqe_buffers_unregister(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008409
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008410 if (ctx->mm_account) {
Jens Axboe2aede0e2020-09-14 10:45:53 -06008411 mmdrop(ctx->mm_account);
8412 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008413 }
Jens Axboedef596e2019-01-09 08:59:42 -07008414
Hao Xu8bad28d2021-02-19 17:19:36 +08008415 mutex_lock(&ctx->uring_lock);
Jens Axboe6b063142019-01-10 22:13:58 -07008416 io_sqe_files_unregister(ctx);
Hao Xu8bad28d2021-02-19 17:19:36 +08008417 mutex_unlock(&ctx->uring_lock);
Jens Axboe9b402842019-04-11 11:45:41 -06008418 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008419 io_destroy_buffers(ctx);
Jens Axboe41726c92020-02-23 13:11:42 -07008420 idr_destroy(&ctx->personality_idr);
Jens Axboedef596e2019-01-09 08:59:42 -07008421
Jens Axboe2b188cc2019-01-07 10:46:33 -07008422#if defined(CONFIG_UNIX)
Eric Biggers355e8d22019-06-12 14:58:43 -07008423 if (ctx->ring_sock) {
8424 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008425 sock_release(ctx->ring_sock);
Eric Biggers355e8d22019-06-12 14:58:43 -07008426 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008427#endif
8428
Hristo Venev75b28af2019-08-26 17:23:46 +00008429 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008430 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008431
8432 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008433 free_uid(ctx->user);
Jens Axboe4010fec2021-02-27 15:04:18 -07008434 io_req_caches_free(ctx);
Jens Axboee9418942021-02-19 12:33:30 -07008435 if (ctx->hash_map)
8436 io_wq_put_hash(ctx->hash_map);
Jens Axboe78076bb2019-12-04 19:56:40 -07008437 kfree(ctx->cancel_hash);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008438 kfree(ctx);
8439}
8440
8441static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8442{
8443 struct io_ring_ctx *ctx = file->private_data;
8444 __poll_t mask = 0;
8445
8446 poll_wait(file, &ctx->cq_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008447 /*
8448 * synchronizes with barrier from wq_has_sleeper call in
8449 * io_commit_cqring
8450 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008451 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008452 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008453 mask |= EPOLLOUT | EPOLLWRNORM;
Hao Xued670c32021-02-05 16:34:21 +08008454
8455 /*
8456 * Don't flush cqring overflow list here, just do a simple check.
8457 * Otherwise there could possible be ABBA deadlock:
8458 * CPU0 CPU1
8459 * ---- ----
8460 * lock(&ctx->uring_lock);
8461 * lock(&ep->mtx);
8462 * lock(&ctx->uring_lock);
8463 * lock(&ep->mtx);
8464 *
8465 * Users may get EPOLLIN meanwhile seeing nothing in cqring, this
8466 * pushs them to do the flush.
8467 */
8468 if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008469 mask |= EPOLLIN | EPOLLRDNORM;
8470
8471 return mask;
8472}
8473
8474static int io_uring_fasync(int fd, struct file *file, int on)
8475{
8476 struct io_ring_ctx *ctx = file->private_data;
8477
8478 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8479}
8480
Yejune Deng0bead8c2020-12-24 11:02:20 +08008481static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
Jens Axboe071698e2020-01-28 10:04:42 -07008482{
Jens Axboe4379bf82021-02-15 13:40:22 -07008483 const struct cred *creds;
Jens Axboe071698e2020-01-28 10:04:42 -07008484
Jens Axboe4379bf82021-02-15 13:40:22 -07008485 creds = idr_remove(&ctx->personality_idr, id);
8486 if (creds) {
8487 put_cred(creds);
Yejune Deng0bead8c2020-12-24 11:02:20 +08008488 return 0;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008489 }
Yejune Deng0bead8c2020-12-24 11:02:20 +08008490
8491 return -EINVAL;
8492}
8493
8494static int io_remove_personalities(int id, void *p, void *data)
8495{
8496 struct io_ring_ctx *ctx = data;
8497
8498 io_unregister_personality(ctx, id);
Jens Axboe071698e2020-01-28 10:04:42 -07008499 return 0;
8500}
8501
Pavel Begunkovba50a032021-02-26 15:47:56 +00008502static bool io_run_ctx_fallback(struct io_ring_ctx *ctx)
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008503{
Pavel Begunkov28c47212021-02-28 22:04:54 +00008504 struct callback_head *work, *next;
Pavel Begunkovba50a032021-02-26 15:47:56 +00008505 bool executed = false;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008506
8507 do {
Pavel Begunkov28c47212021-02-28 22:04:54 +00008508 work = xchg(&ctx->exit_task_work, NULL);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008509 if (!work)
8510 break;
8511
8512 do {
8513 next = work->next;
8514 work->func(work);
8515 work = next;
8516 cond_resched();
8517 } while (work);
Pavel Begunkovba50a032021-02-26 15:47:56 +00008518 executed = true;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008519 } while (1);
Pavel Begunkovba50a032021-02-26 15:47:56 +00008520
8521 return executed;
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008522}
8523
Jens Axboe85faa7b2020-04-09 18:14:00 -06008524static void io_ring_exit_work(struct work_struct *work)
8525{
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008526 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx,
8527 exit_work);
Jens Axboe85faa7b2020-04-09 18:14:00 -06008528
Jens Axboe56952e92020-06-17 15:00:04 -06008529 /*
8530 * If we're doing polled IO and end up having requests being
8531 * submitted async (out-of-line), then completions can come in while
8532 * we're waiting for refs to drop. We need to reap these manually,
8533 * as nobody else will be looking for them.
8534 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008535 do {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008536 io_uring_try_cancel_requests(ctx, NULL, NULL);
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008537 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Jens Axboe85faa7b2020-04-09 18:14:00 -06008538 io_ring_ctx_free(ctx);
8539}
8540
Jens Axboe2b188cc2019-01-07 10:46:33 -07008541static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8542{
8543 mutex_lock(&ctx->uring_lock);
8544 percpu_ref_kill(&ctx->refs);
Pavel Begunkovcda286f2020-12-17 00:24:35 +00008545 /* if force is set, the ring is going away. always drop after that */
8546 ctx->cq_overflow_flushed = 1;
Pavel Begunkov634578f2020-12-06 22:22:44 +00008547 if (ctx->rings)
Pavel Begunkov6c503152021-01-04 20:36:36 +00008548 __io_cqring_overflow_flush(ctx, true, NULL, NULL);
Pavel Begunkov5c766a92021-01-19 13:32:36 +00008549 idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008550 mutex_unlock(&ctx->uring_lock);
8551
Pavel Begunkov6b819282020-11-06 13:00:25 +00008552 io_kill_timeouts(ctx, NULL, NULL);
8553 io_poll_remove_all(ctx, NULL, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06008554
Jens Axboe15dff282019-11-13 09:09:23 -07008555 /* if we failed setting up the ctx, we might not have any rings */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008556 io_iopoll_try_reap_events(ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008557
Jens Axboe85faa7b2020-04-09 18:14:00 -06008558 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008559 /*
8560 * Use system_unbound_wq to avoid spawning tons of event kworkers
8561 * if we're exiting a ton of rings at the same time. It just adds
8562 * noise and overhead, there's no discernable change in runtime
8563 * over using system_wq.
8564 */
8565 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008566}
8567
8568static int io_uring_release(struct inode *inode, struct file *file)
8569{
8570 struct io_ring_ctx *ctx = file->private_data;
8571
8572 file->private_data = NULL;
8573 io_ring_ctx_wait_and_kill(ctx);
8574 return 0;
8575}
8576
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008577struct io_task_cancel {
8578 struct task_struct *task;
8579 struct files_struct *files;
8580};
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008581
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008582static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Jens Axboeb711d4e2020-08-16 08:23:05 -07008583{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008584 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008585 struct io_task_cancel *cancel = data;
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008586 bool ret;
8587
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008588 if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) {
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008589 unsigned long flags;
8590 struct io_ring_ctx *ctx = req->ctx;
8591
8592 /* protect against races with linked timeouts */
8593 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008594 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008595 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8596 } else {
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008597 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008598 }
8599 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008600}
8601
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008602static void io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008603 struct task_struct *task,
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008604 struct files_struct *files)
8605{
8606 struct io_defer_entry *de = NULL;
8607 LIST_HEAD(list);
8608
8609 spin_lock_irq(&ctx->completion_lock);
8610 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00008611 if (io_match_task(de->req, task, files)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008612 list_cut_position(&list, &ctx->defer_list, &de->list);
8613 break;
8614 }
8615 }
8616 spin_unlock_irq(&ctx->completion_lock);
8617
8618 while (!list_empty(&list)) {
8619 de = list_first_entry(&list, struct io_defer_entry, list);
8620 list_del_init(&de->list);
8621 req_set_fail_links(de->req);
8622 io_put_req(de->req);
8623 io_req_complete(de->req, -ECANCELED);
8624 kfree(de);
8625 }
8626}
8627
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008628static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
8629 struct task_struct *task,
8630 struct files_struct *files)
8631{
8632 struct io_task_cancel cancel = { .task = task, .files = files, };
Pavel Begunkov64c72122021-03-01 18:20:45 +00008633 struct task_struct *tctx_task = task ?: current;
8634 struct io_uring_task *tctx = tctx_task->io_uring;
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008635
8636 while (1) {
8637 enum io_wq_cancel cret;
8638 bool ret = false;
8639
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008640 if (tctx && tctx->io_wq) {
8641 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb,
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008642 &cancel, true);
8643 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8644 }
8645
8646 /* SQPOLL thread does its own polling */
8647 if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) {
8648 while (!list_empty_careful(&ctx->iopoll_list)) {
8649 io_iopoll_try_reap_events(ctx);
8650 ret = true;
8651 }
8652 }
8653
8654 ret |= io_poll_remove_all(ctx, task, files);
8655 ret |= io_kill_timeouts(ctx, task, files);
8656 ret |= io_run_task_work();
Pavel Begunkovba50a032021-02-26 15:47:56 +00008657 ret |= io_run_ctx_fallback(ctx);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008658 io_cqring_overflow_flush(ctx, true, task, files);
8659 if (!ret)
8660 break;
8661 cond_resched();
8662 }
8663}
8664
Pavel Begunkovca70f002021-01-26 15:28:27 +00008665static int io_uring_count_inflight(struct io_ring_ctx *ctx,
8666 struct task_struct *task,
8667 struct files_struct *files)
8668{
8669 struct io_kiocb *req;
8670 int cnt = 0;
8671
8672 spin_lock_irq(&ctx->inflight_lock);
8673 list_for_each_entry(req, &ctx->inflight_list, inflight_entry)
8674 cnt += io_match_task(req, task, files);
8675 spin_unlock_irq(&ctx->inflight_lock);
8676 return cnt;
8677}
8678
Pavel Begunkovb52fda02020-11-06 13:00:24 +00008679static void io_uring_cancel_files(struct io_ring_ctx *ctx,
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00008680 struct task_struct *task,
Jens Axboefcb323c2019-10-24 12:39:47 -06008681 struct files_struct *files)
8682{
Jens Axboefcb323c2019-10-24 12:39:47 -06008683 while (!list_empty_careful(&ctx->inflight_list)) {
Xiaoguang Wangd8f1b972020-04-26 15:54:43 +08008684 DEFINE_WAIT(wait);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008685 int inflight;
Jens Axboefcb323c2019-10-24 12:39:47 -06008686
Pavel Begunkovca70f002021-01-26 15:28:27 +00008687 inflight = io_uring_count_inflight(ctx, task, files);
8688 if (!inflight)
Jens Axboefcb323c2019-10-24 12:39:47 -06008689 break;
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008690
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008691 io_uring_try_cancel_requests(ctx, task, files);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008692
Pavel Begunkov34343782021-02-10 11:45:42 +00008693 if (ctx->sq_data)
8694 io_sq_thread_unpark(ctx->sq_data);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008695 prepare_to_wait(&task->io_uring->wait, &wait,
8696 TASK_UNINTERRUPTIBLE);
8697 if (inflight == io_uring_count_inflight(ctx, task, files))
8698 schedule();
Pavel Begunkovc98de082020-11-15 12:56:32 +00008699 finish_wait(&task->io_uring->wait, &wait);
Pavel Begunkov34343782021-02-10 11:45:42 +00008700 if (ctx->sq_data)
8701 io_sq_thread_park(ctx->sq_data);
Jens Axboe0f212202020-09-13 13:09:39 -06008702 }
Jens Axboe0f212202020-09-13 13:09:39 -06008703}
8704
8705/*
8706 * We need to iteratively cancel requests, in case a request has dependent
8707 * hard links. These persist even for failure of cancelations, hence keep
8708 * looping until none are found.
8709 */
8710static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
8711 struct files_struct *files)
8712{
8713 struct task_struct *task = current;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008714 bool did_park = false;
Jens Axboe0f212202020-09-13 13:09:39 -06008715
Jens Axboefdaf0832020-10-30 09:37:30 -06008716 if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00008717 /* never started, nothing to cancel */
8718 if (ctx->flags & IORING_SETUP_R_DISABLED) {
8719 io_sq_offload_start(ctx);
8720 return;
8721 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008722 did_park = io_sq_thread_park(ctx->sq_data);
8723 if (did_park) {
8724 task = ctx->sq_data->thread;
8725 atomic_inc(&task->io_uring->in_idle);
8726 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008727 }
Jens Axboe0f212202020-09-13 13:09:39 -06008728
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00008729 io_cancel_defer_files(ctx, task, files);
Jens Axboe0f212202020-09-13 13:09:39 -06008730
Pavel Begunkov3a7efd12021-01-28 23:23:42 +00008731 io_uring_cancel_files(ctx, task, files);
Pavel Begunkovb52fda02020-11-06 13:00:24 +00008732 if (!files)
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008733 io_uring_try_cancel_requests(ctx, task, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06008734
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008735 if (did_park) {
Jens Axboefdaf0832020-10-30 09:37:30 -06008736 atomic_dec(&task->io_uring->in_idle);
Jens Axboefdaf0832020-10-30 09:37:30 -06008737 io_sq_thread_unpark(ctx->sq_data);
8738 }
Jens Axboe0f212202020-09-13 13:09:39 -06008739}
8740
8741/*
8742 * Note that this task has used io_uring. We use it for cancelation purposes.
8743 */
Jens Axboefdaf0832020-10-30 09:37:30 -06008744static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file)
Jens Axboe0f212202020-09-13 13:09:39 -06008745{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008746 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkova528b042020-12-21 18:34:04 +00008747 int ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008748
8749 if (unlikely(!tctx)) {
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008750 ret = io_uring_alloc_task_context(current, ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06008751 if (unlikely(ret))
8752 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008753 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008754 }
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008755 if (tctx->last != file) {
8756 void *old = xa_load(&tctx->xa, (unsigned long)file);
Jens Axboe0f212202020-09-13 13:09:39 -06008757
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008758 if (!old) {
Jens Axboe0f212202020-09-13 13:09:39 -06008759 get_file(file);
Pavel Begunkova528b042020-12-21 18:34:04 +00008760 ret = xa_err(xa_store(&tctx->xa, (unsigned long)file,
8761 file, GFP_KERNEL));
8762 if (ret) {
8763 fput(file);
8764 return ret;
8765 }
Jens Axboe0f212202020-09-13 13:09:39 -06008766 }
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008767 tctx->last = file;
Jens Axboe0f212202020-09-13 13:09:39 -06008768 }
8769
Jens Axboefdaf0832020-10-30 09:37:30 -06008770 /*
8771 * This is race safe in that the task itself is doing this, hence it
8772 * cannot be going through the exit/cancel paths at the same time.
8773 * This cannot be modified while exit/cancel is running.
8774 */
8775 if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL))
8776 tctx->sqpoll = true;
8777
Jens Axboe0f212202020-09-13 13:09:39 -06008778 return 0;
8779}
8780
8781/*
8782 * Remove this io_uring_file -> task mapping.
8783 */
8784static void io_uring_del_task_file(struct file *file)
8785{
8786 struct io_uring_task *tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008787
8788 if (tctx->last == file)
8789 tctx->last = NULL;
Matthew Wilcox (Oracle)5e2ed8c2020-10-09 13:49:53 +01008790 file = xa_erase(&tctx->xa, (unsigned long)file);
Jens Axboe0f212202020-09-13 13:09:39 -06008791 if (file)
8792 fput(file);
8793}
8794
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008795static void io_uring_clean_tctx(struct io_uring_task *tctx)
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008796{
8797 struct file *file;
8798 unsigned long index;
8799
8800 xa_for_each(&tctx->xa, index, file)
8801 io_uring_del_task_file(file);
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008802 if (tctx->io_wq) {
8803 io_wq_put_and_exit(tctx->io_wq);
8804 tctx->io_wq = NULL;
8805 }
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008806}
8807
Jens Axboe0f212202020-09-13 13:09:39 -06008808void __io_uring_files_cancel(struct files_struct *files)
8809{
8810 struct io_uring_task *tctx = current->io_uring;
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008811 struct file *file;
8812 unsigned long index;
Jens Axboe0f212202020-09-13 13:09:39 -06008813
8814 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008815 atomic_inc(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008816 xa_for_each(&tctx->xa, index, file)
8817 io_uring_cancel_task_requests(file->private_data, files);
Jens Axboefdaf0832020-10-30 09:37:30 -06008818 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008819
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008820 if (files)
8821 io_uring_clean_tctx(tctx);
Jens Axboefdaf0832020-10-30 09:37:30 -06008822}
8823
8824static s64 tctx_inflight(struct io_uring_task *tctx)
8825{
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008826 return percpu_counter_sum(&tctx->inflight);
8827}
8828
8829static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx)
8830{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008831 struct io_sq_data *sqd = ctx->sq_data;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008832 struct io_uring_task *tctx;
Jens Axboefdaf0832020-10-30 09:37:30 -06008833 s64 inflight;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008834 DEFINE_WAIT(wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06008835
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008836 if (!sqd)
8837 return;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008838 if (!io_sq_thread_park(sqd))
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008839 return;
8840 tctx = ctx->sq_data->thread->io_uring;
Jens Axboee54945a2021-02-26 11:27:15 -07008841 /* can happen on fork/alloc failure, just ignore that state */
8842 if (!tctx) {
8843 io_sq_thread_unpark(sqd);
8844 return;
8845 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008846
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008847 atomic_inc(&tctx->in_idle);
8848 do {
8849 /* read completions before cancelations */
8850 inflight = tctx_inflight(tctx);
8851 if (!inflight)
8852 break;
8853 io_uring_cancel_task_requests(ctx, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06008854
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008855 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8856 /*
8857 * If we've seen completions, retry without waiting. This
8858 * avoids a race where a completion comes in before we did
8859 * prepare_to_wait().
8860 */
8861 if (inflight == tctx_inflight(tctx))
8862 schedule();
8863 finish_wait(&tctx->wait, &wait);
8864 } while (1);
8865 atomic_dec(&tctx->in_idle);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008866 io_sq_thread_unpark(sqd);
Jens Axboe0f212202020-09-13 13:09:39 -06008867}
8868
Jens Axboe0f212202020-09-13 13:09:39 -06008869/*
8870 * Find any io_uring fd that this task has registered or done IO on, and cancel
8871 * requests.
8872 */
8873void __io_uring_task_cancel(void)
8874{
8875 struct io_uring_task *tctx = current->io_uring;
8876 DEFINE_WAIT(wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008877 s64 inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06008878
8879 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008880 atomic_inc(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06008881
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008882 if (tctx->sqpoll) {
8883 struct file *file;
8884 unsigned long index;
8885
8886 xa_for_each(&tctx->xa, index, file)
8887 io_uring_cancel_sqpoll(file->private_data);
8888 }
Pavel Begunkov0b5cd6c2021-01-17 02:29:56 +00008889
Jens Axboed8a6df12020-10-15 16:24:45 -06008890 do {
Jens Axboe0f212202020-09-13 13:09:39 -06008891 /* read completions before cancelations */
Jens Axboefdaf0832020-10-30 09:37:30 -06008892 inflight = tctx_inflight(tctx);
Jens Axboed8a6df12020-10-15 16:24:45 -06008893 if (!inflight)
8894 break;
Jens Axboe0f212202020-09-13 13:09:39 -06008895 __io_uring_files_cancel(NULL);
8896
8897 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8898
8899 /*
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00008900 * If we've seen completions, retry without waiting. This
8901 * avoids a race where a completion comes in before we did
8902 * prepare_to_wait().
Jens Axboe0f212202020-09-13 13:09:39 -06008903 */
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00008904 if (inflight == tctx_inflight(tctx))
8905 schedule();
Pavel Begunkovf57555e2020-12-20 13:21:44 +00008906 finish_wait(&tctx->wait, &wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008907 } while (1);
Jens Axboe0f212202020-09-13 13:09:39 -06008908
Jens Axboefdaf0832020-10-30 09:37:30 -06008909 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008910
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008911 io_uring_clean_tctx(tctx);
8912 /* all current's requests should be gone, we can kill tctx */
8913 __io_uring_free(current);
Pavel Begunkov44e728b2020-06-15 10:24:04 +03008914}
8915
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008916static void *io_uring_validate_mmap_request(struct file *file,
8917 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008918{
Jens Axboe2b188cc2019-01-07 10:46:33 -07008919 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008920 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008921 struct page *page;
8922 void *ptr;
8923
8924 switch (offset) {
8925 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00008926 case IORING_OFF_CQ_RING:
8927 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008928 break;
8929 case IORING_OFF_SQES:
8930 ptr = ctx->sq_sqes;
8931 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008932 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008933 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008934 }
8935
8936 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07008937 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008938 return ERR_PTR(-EINVAL);
8939
8940 return ptr;
8941}
8942
8943#ifdef CONFIG_MMU
8944
8945static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
8946{
8947 size_t sz = vma->vm_end - vma->vm_start;
8948 unsigned long pfn;
8949 void *ptr;
8950
8951 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
8952 if (IS_ERR(ptr))
8953 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008954
8955 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
8956 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
8957}
8958
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008959#else /* !CONFIG_MMU */
8960
8961static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
8962{
8963 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
8964}
8965
8966static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
8967{
8968 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
8969}
8970
8971static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
8972 unsigned long addr, unsigned long len,
8973 unsigned long pgoff, unsigned long flags)
8974{
8975 void *ptr;
8976
8977 ptr = io_uring_validate_mmap_request(file, pgoff, len);
8978 if (IS_ERR(ptr))
8979 return PTR_ERR(ptr);
8980
8981 return (unsigned long) ptr;
8982}
8983
8984#endif /* !CONFIG_MMU */
8985
Pavel Begunkovd9d05212021-01-08 20:57:25 +00008986static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
Jens Axboe90554202020-09-03 12:12:41 -06008987{
Pavel Begunkovd9d05212021-01-08 20:57:25 +00008988 int ret = 0;
Jens Axboe90554202020-09-03 12:12:41 -06008989 DEFINE_WAIT(wait);
8990
8991 do {
8992 if (!io_sqring_full(ctx))
8993 break;
Jens Axboe90554202020-09-03 12:12:41 -06008994 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
8995
8996 if (!io_sqring_full(ctx))
8997 break;
Jens Axboe90554202020-09-03 12:12:41 -06008998 schedule();
8999 } while (!signal_pending(current));
9000
9001 finish_wait(&ctx->sqo_sq_wait, &wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009002 return ret;
Jens Axboe90554202020-09-03 12:12:41 -06009003}
9004
Hao Xuc73ebb62020-11-03 10:54:37 +08009005static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
9006 struct __kernel_timespec __user **ts,
9007 const sigset_t __user **sig)
9008{
9009 struct io_uring_getevents_arg arg;
9010
9011 /*
9012 * If EXT_ARG isn't set, then we have no timespec and the argp pointer
9013 * is just a pointer to the sigset_t.
9014 */
9015 if (!(flags & IORING_ENTER_EXT_ARG)) {
9016 *sig = (const sigset_t __user *) argp;
9017 *ts = NULL;
9018 return 0;
9019 }
9020
9021 /*
9022 * EXT_ARG is set - ensure we agree on the size of it and copy in our
9023 * timespec and sigset_t pointers if good.
9024 */
9025 if (*argsz != sizeof(arg))
9026 return -EINVAL;
9027 if (copy_from_user(&arg, argp, sizeof(arg)))
9028 return -EFAULT;
9029 *sig = u64_to_user_ptr(arg.sigmask);
9030 *argsz = arg.sigmask_sz;
9031 *ts = u64_to_user_ptr(arg.ts);
9032 return 0;
9033}
9034
Jens Axboe2b188cc2019-01-07 10:46:33 -07009035SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
Hao Xuc73ebb62020-11-03 10:54:37 +08009036 u32, min_complete, u32, flags, const void __user *, argp,
9037 size_t, argsz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009038{
9039 struct io_ring_ctx *ctx;
9040 long ret = -EBADF;
9041 int submitted = 0;
9042 struct fd f;
9043
Jens Axboe4c6e2772020-07-01 11:29:10 -06009044 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07009045
Jens Axboe90554202020-09-03 12:12:41 -06009046 if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
Hao Xuc73ebb62020-11-03 10:54:37 +08009047 IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009048 return -EINVAL;
9049
9050 f = fdget(fd);
9051 if (!f.file)
9052 return -EBADF;
9053
9054 ret = -EOPNOTSUPP;
9055 if (f.file->f_op != &io_uring_fops)
9056 goto out_fput;
9057
9058 ret = -ENXIO;
9059 ctx = f.file->private_data;
9060 if (!percpu_ref_tryget(&ctx->refs))
9061 goto out_fput;
9062
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009063 ret = -EBADFD;
9064 if (ctx->flags & IORING_SETUP_R_DISABLED)
9065 goto out;
9066
Jens Axboe6c271ce2019-01-10 11:22:30 -07009067 /*
9068 * For SQ polling, the thread will do all submissions and completions.
9069 * Just return the requested submit count, and wake the thread if
9070 * we were asked to.
9071 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009072 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009073 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00009074 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Pavel Begunkov89448c42020-12-17 00:24:39 +00009075
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009076 if (unlikely(ctx->sqo_exec)) {
9077 ret = io_sq_thread_fork(ctx->sq_data, ctx);
9078 if (ret)
9079 goto out;
9080 ctx->sqo_exec = 0;
9081 }
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009082 ret = -EOWNERDEAD;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009083 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06009084 wake_up(&ctx->sq_data->wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009085 if (flags & IORING_ENTER_SQ_WAIT) {
9086 ret = io_sqpoll_wait_sq(ctx);
9087 if (ret)
9088 goto out;
9089 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009090 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009091 } else if (to_submit) {
Jens Axboefdaf0832020-10-30 09:37:30 -06009092 ret = io_uring_add_task_file(ctx, f.file);
Jens Axboe0f212202020-09-13 13:09:39 -06009093 if (unlikely(ret))
9094 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009095 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009096 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009097 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009098
9099 if (submitted != to_submit)
9100 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009101 }
9102 if (flags & IORING_ENTER_GETEVENTS) {
Hao Xuc73ebb62020-11-03 10:54:37 +08009103 const sigset_t __user *sig;
9104 struct __kernel_timespec __user *ts;
9105
9106 ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
9107 if (unlikely(ret))
9108 goto out;
9109
Jens Axboe2b188cc2019-01-07 10:46:33 -07009110 min_complete = min(min_complete, ctx->cq_entries);
9111
Xiaoguang Wang32b22442020-03-11 09:26:09 +08009112 /*
9113 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
9114 * space applications don't need to do io completion events
9115 * polling again, they can rely on io_sq_thread to do polling
9116 * work, which can reduce cpu usage and uring_lock contention.
9117 */
9118 if (ctx->flags & IORING_SETUP_IOPOLL &&
9119 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03009120 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07009121 } else {
Hao Xuc73ebb62020-11-03 10:54:37 +08009122 ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts);
Jens Axboedef596e2019-01-09 08:59:42 -07009123 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009124 }
9125
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009126out:
Pavel Begunkov6805b322019-10-08 02:18:42 +03009127 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009128out_fput:
9129 fdput(f);
9130 return submitted ? submitted : ret;
9131}
9132
Tobias Klauserbebdb652020-02-26 18:38:32 +01009133#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009134static int io_uring_show_cred(int id, void *p, void *data)
9135{
Jens Axboe4379bf82021-02-15 13:40:22 -07009136 const struct cred *cred = p;
Jens Axboe87ce9552020-01-30 08:25:34 -07009137 struct seq_file *m = data;
9138 struct user_namespace *uns = seq_user_ns(m);
9139 struct group_info *gi;
9140 kernel_cap_t cap;
9141 unsigned __capi;
9142 int g;
9143
9144 seq_printf(m, "%5d\n", id);
9145 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9146 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9147 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9148 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9149 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9150 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9151 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9152 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9153 seq_puts(m, "\n\tGroups:\t");
9154 gi = cred->group_info;
9155 for (g = 0; g < gi->ngroups; g++) {
9156 seq_put_decimal_ull(m, g ? " " : "",
9157 from_kgid_munged(uns, gi->gid[g]));
9158 }
9159 seq_puts(m, "\n\tCapEff:\t");
9160 cap = cred->cap_effective;
9161 CAP_FOR_EACH_U32(__capi)
9162 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9163 seq_putc(m, '\n');
9164 return 0;
9165}
9166
9167static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9168{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009169 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009170 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009171 int i;
9172
Jens Axboefad8e0d2020-09-28 08:57:48 -06009173 /*
9174 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9175 * since fdinfo case grabs it in the opposite direction of normal use
9176 * cases. If we fail to get the lock, we just don't iterate any
9177 * structures that could be going away outside the io_uring mutex.
9178 */
9179 has_lock = mutex_trylock(&ctx->uring_lock);
9180
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009181 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
Joseph Qidbbe9c62020-09-29 09:01:22 -06009182 sq = ctx->sq_data;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009183 if (!sq->thread)
9184 sq = NULL;
9185 }
Joseph Qidbbe9c62020-09-29 09:01:22 -06009186
9187 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9188 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009189 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009190 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Pavel Begunkovea64ec022021-02-04 13:52:07 +00009191 struct file *f = *io_fixed_file_slot(ctx->file_data, i);
Jens Axboe87ce9552020-01-30 08:25:34 -07009192
Jens Axboe87ce9552020-01-30 08:25:34 -07009193 if (f)
9194 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9195 else
9196 seq_printf(m, "%5u: <none>\n", i);
9197 }
9198 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009199 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009200 struct io_mapped_ubuf *buf = &ctx->user_bufs[i];
9201
9202 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf,
9203 (unsigned int) buf->len);
9204 }
Jens Axboefad8e0d2020-09-28 08:57:48 -06009205 if (has_lock && !idr_is_empty(&ctx->personality_idr)) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009206 seq_printf(m, "Personalities:\n");
9207 idr_for_each(&ctx->personality_idr, io_uring_show_cred, m);
9208 }
Jens Axboed7718a92020-02-14 22:23:12 -07009209 seq_printf(m, "PollList:\n");
9210 spin_lock_irq(&ctx->completion_lock);
9211 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9212 struct hlist_head *list = &ctx->cancel_hash[i];
9213 struct io_kiocb *req;
9214
9215 hlist_for_each_entry(req, list, hash_node)
9216 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9217 req->task->task_works != NULL);
9218 }
9219 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009220 if (has_lock)
9221 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009222}
9223
9224static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9225{
9226 struct io_ring_ctx *ctx = f->private_data;
9227
9228 if (percpu_ref_tryget(&ctx->refs)) {
9229 __io_uring_show_fdinfo(ctx, m);
9230 percpu_ref_put(&ctx->refs);
9231 }
9232}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009233#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009234
Jens Axboe2b188cc2019-01-07 10:46:33 -07009235static const struct file_operations io_uring_fops = {
9236 .release = io_uring_release,
9237 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009238#ifndef CONFIG_MMU
9239 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9240 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9241#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009242 .poll = io_uring_poll,
9243 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009244#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009245 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009246#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009247};
9248
9249static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9250 struct io_uring_params *p)
9251{
Hristo Venev75b28af2019-08-26 17:23:46 +00009252 struct io_rings *rings;
9253 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009254
Jens Axboebd740482020-08-05 12:58:23 -06009255 /* make sure these are sane, as we already accounted them */
9256 ctx->sq_entries = p->sq_entries;
9257 ctx->cq_entries = p->cq_entries;
9258
Hristo Venev75b28af2019-08-26 17:23:46 +00009259 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9260 if (size == SIZE_MAX)
9261 return -EOVERFLOW;
9262
9263 rings = io_mem_alloc(size);
9264 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009265 return -ENOMEM;
9266
Hristo Venev75b28af2019-08-26 17:23:46 +00009267 ctx->rings = rings;
9268 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9269 rings->sq_ring_mask = p->sq_entries - 1;
9270 rings->cq_ring_mask = p->cq_entries - 1;
9271 rings->sq_ring_entries = p->sq_entries;
9272 rings->cq_ring_entries = p->cq_entries;
9273 ctx->sq_mask = rings->sq_ring_mask;
9274 ctx->cq_mask = rings->cq_ring_mask;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009275
9276 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009277 if (size == SIZE_MAX) {
9278 io_mem_free(ctx->rings);
9279 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009280 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009281 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009282
9283 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009284 if (!ctx->sq_sqes) {
9285 io_mem_free(ctx->rings);
9286 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009287 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009288 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009289
Jens Axboe2b188cc2019-01-07 10:46:33 -07009290 return 0;
9291}
9292
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009293static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)
9294{
9295 int ret, fd;
9296
9297 fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9298 if (fd < 0)
9299 return fd;
9300
9301 ret = io_uring_add_task_file(ctx, file);
9302 if (ret) {
9303 put_unused_fd(fd);
9304 return ret;
9305 }
9306 fd_install(fd, file);
9307 return fd;
9308}
9309
Jens Axboe2b188cc2019-01-07 10:46:33 -07009310/*
9311 * Allocate an anonymous fd, this is what constitutes the application
9312 * visible backing of an io_uring instance. The application mmaps this
9313 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9314 * we have to tie this fd to a socket for file garbage collection purposes.
9315 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009316static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009317{
9318 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009319#if defined(CONFIG_UNIX)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009320 int ret;
9321
Jens Axboe2b188cc2019-01-07 10:46:33 -07009322 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9323 &ctx->ring_sock);
9324 if (ret)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009325 return ERR_PTR(ret);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009326#endif
9327
Jens Axboe2b188cc2019-01-07 10:46:33 -07009328 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9329 O_RDWR | O_CLOEXEC);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009330#if defined(CONFIG_UNIX)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009331 if (IS_ERR(file)) {
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009332 sock_release(ctx->ring_sock);
9333 ctx->ring_sock = NULL;
9334 } else {
9335 ctx->ring_sock->file = file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009336 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009337#endif
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009338 return file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009339}
9340
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009341static int io_uring_create(unsigned entries, struct io_uring_params *p,
9342 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009343{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009344 struct io_ring_ctx *ctx;
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009345 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009346 int ret;
9347
Jens Axboe8110c1a2019-12-28 15:39:54 -07009348 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009349 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009350 if (entries > IORING_MAX_ENTRIES) {
9351 if (!(p->flags & IORING_SETUP_CLAMP))
9352 return -EINVAL;
9353 entries = IORING_MAX_ENTRIES;
9354 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009355
9356 /*
9357 * Use twice as many entries for the CQ ring. It's possible for the
9358 * application to drive a higher depth than the size of the SQ ring,
9359 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009360 * some flexibility in overcommitting a bit. If the application has
9361 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9362 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009363 */
9364 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009365 if (p->flags & IORING_SETUP_CQSIZE) {
9366 /*
9367 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9368 * to a power-of-two, if it isn't already. We do NOT impose
9369 * any cq vs sq ring sizing.
9370 */
Joseph Qieb2667b32020-11-24 15:03:03 +08009371 if (!p->cq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009372 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009373 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9374 if (!(p->flags & IORING_SETUP_CLAMP))
9375 return -EINVAL;
9376 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9377 }
Joseph Qieb2667b32020-11-24 15:03:03 +08009378 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9379 if (p->cq_entries < p->sq_entries)
9380 return -EINVAL;
Jens Axboe33a107f2019-10-04 12:10:03 -06009381 } else {
9382 p->cq_entries = 2 * p->sq_entries;
9383 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009384
Jens Axboe2b188cc2019-01-07 10:46:33 -07009385 ctx = io_ring_ctx_alloc(p);
Jens Axboe62e398b2021-02-21 16:19:37 -07009386 if (!ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009387 return -ENOMEM;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009388 ctx->compat = in_compat_syscall();
Jens Axboe62e398b2021-02-21 16:19:37 -07009389 if (!capable(CAP_IPC_LOCK))
9390 ctx->user = get_uid(current_user());
Jens Axboe2aede0e2020-09-14 10:45:53 -06009391
9392 /*
9393 * This is just grabbed for accounting purposes. When a process exits,
9394 * the mm is exited and dropped before the files, hence we need to hang
9395 * on to this mm purely for the purposes of being able to unaccount
9396 * memory (locked/pinned vm). It's not used for anything else.
9397 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009398 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009399 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009400
Jens Axboe2b188cc2019-01-07 10:46:33 -07009401 ret = io_allocate_scq_urings(ctx, p);
9402 if (ret)
9403 goto err;
9404
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009405 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009406 if (ret)
9407 goto err;
9408
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009409 if (!(p->flags & IORING_SETUP_R_DISABLED))
9410 io_sq_offload_start(ctx);
9411
Jens Axboe2b188cc2019-01-07 10:46:33 -07009412 memset(&p->sq_off, 0, sizeof(p->sq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009413 p->sq_off.head = offsetof(struct io_rings, sq.head);
9414 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9415 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9416 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9417 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9418 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9419 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009420
9421 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009422 p->cq_off.head = offsetof(struct io_rings, cq.head);
9423 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9424 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9425 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9426 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9427 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009428 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009429
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009430 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9431 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009432 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
Hao Xuc73ebb62020-11-03 10:54:37 +08009433 IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
Jens Axboe1c0aa1f2021-02-20 11:55:28 -07009434 IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009435
9436 if (copy_to_user(params, p, sizeof(*p))) {
9437 ret = -EFAULT;
9438 goto err;
9439 }
Jens Axboed1719f72020-07-30 13:43:53 -06009440
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009441 file = io_uring_get_file(ctx);
9442 if (IS_ERR(file)) {
9443 ret = PTR_ERR(file);
9444 goto err;
9445 }
9446
Jens Axboed1719f72020-07-30 13:43:53 -06009447 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009448 * Install ring fd as the very last thing, so we don't risk someone
9449 * having closed it before we finish setup
9450 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009451 ret = io_uring_install_fd(ctx, file);
9452 if (ret < 0) {
9453 /* fput will clean it up */
9454 fput(file);
9455 return ret;
9456 }
Jens Axboe044c1ab2019-10-28 09:15:33 -06009457
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009458 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009459 return ret;
9460err:
9461 io_ring_ctx_wait_and_kill(ctx);
9462 return ret;
9463}
9464
9465/*
9466 * Sets up an aio uring context, and returns the fd. Applications asks for a
9467 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9468 * params structure passed in.
9469 */
9470static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9471{
9472 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009473 int i;
9474
9475 if (copy_from_user(&p, params, sizeof(p)))
9476 return -EFAULT;
9477 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9478 if (p.resv[i])
9479 return -EINVAL;
9480 }
9481
Jens Axboe6c271ce2019-01-10 11:22:30 -07009482 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009483 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009484 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9485 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009486 return -EINVAL;
9487
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009488 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009489}
9490
9491SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9492 struct io_uring_params __user *, params)
9493{
9494 return io_uring_setup(entries, params);
9495}
9496
Jens Axboe66f4af92020-01-16 15:36:52 -07009497static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9498{
9499 struct io_uring_probe *p;
9500 size_t size;
9501 int i, ret;
9502
9503 size = struct_size(p, ops, nr_args);
9504 if (size == SIZE_MAX)
9505 return -EOVERFLOW;
9506 p = kzalloc(size, GFP_KERNEL);
9507 if (!p)
9508 return -ENOMEM;
9509
9510 ret = -EFAULT;
9511 if (copy_from_user(p, arg, size))
9512 goto out;
9513 ret = -EINVAL;
9514 if (memchr_inv(p, 0, size))
9515 goto out;
9516
9517 p->last_op = IORING_OP_LAST - 1;
9518 if (nr_args > IORING_OP_LAST)
9519 nr_args = IORING_OP_LAST;
9520
9521 for (i = 0; i < nr_args; i++) {
9522 p->ops[i].op = i;
9523 if (!io_op_defs[i].not_supported)
9524 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9525 }
9526 p->ops_len = i;
9527
9528 ret = 0;
9529 if (copy_to_user(arg, p, size))
9530 ret = -EFAULT;
9531out:
9532 kfree(p);
9533 return ret;
9534}
9535
Jens Axboe071698e2020-01-28 10:04:42 -07009536static int io_register_personality(struct io_ring_ctx *ctx)
9537{
Jens Axboe4379bf82021-02-15 13:40:22 -07009538 const struct cred *creds;
Jens Axboe1e6fa522020-10-15 08:46:24 -06009539 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009540
Jens Axboe4379bf82021-02-15 13:40:22 -07009541 creds = get_current_cred();
Jens Axboe1e6fa522020-10-15 08:46:24 -06009542
Jens Axboe4379bf82021-02-15 13:40:22 -07009543 ret = idr_alloc_cyclic(&ctx->personality_idr, (void *) creds, 1,
9544 USHRT_MAX, GFP_KERNEL);
9545 if (ret < 0)
9546 put_cred(creds);
Jens Axboe1e6fa522020-10-15 08:46:24 -06009547 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009548}
9549
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009550static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9551 unsigned int nr_args)
9552{
9553 struct io_uring_restriction *res;
9554 size_t size;
9555 int i, ret;
9556
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009557 /* Restrictions allowed only if rings started disabled */
9558 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9559 return -EBADFD;
9560
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009561 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009562 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009563 return -EBUSY;
9564
9565 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9566 return -EINVAL;
9567
9568 size = array_size(nr_args, sizeof(*res));
9569 if (size == SIZE_MAX)
9570 return -EOVERFLOW;
9571
9572 res = memdup_user(arg, size);
9573 if (IS_ERR(res))
9574 return PTR_ERR(res);
9575
9576 ret = 0;
9577
9578 for (i = 0; i < nr_args; i++) {
9579 switch (res[i].opcode) {
9580 case IORING_RESTRICTION_REGISTER_OP:
9581 if (res[i].register_op >= IORING_REGISTER_LAST) {
9582 ret = -EINVAL;
9583 goto out;
9584 }
9585
9586 __set_bit(res[i].register_op,
9587 ctx->restrictions.register_op);
9588 break;
9589 case IORING_RESTRICTION_SQE_OP:
9590 if (res[i].sqe_op >= IORING_OP_LAST) {
9591 ret = -EINVAL;
9592 goto out;
9593 }
9594
9595 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9596 break;
9597 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9598 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9599 break;
9600 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9601 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9602 break;
9603 default:
9604 ret = -EINVAL;
9605 goto out;
9606 }
9607 }
9608
9609out:
9610 /* Reset all restrictions if an error happened */
9611 if (ret != 0)
9612 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9613 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009614 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009615
9616 kfree(res);
9617 return ret;
9618}
9619
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009620static int io_register_enable_rings(struct io_ring_ctx *ctx)
9621{
9622 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9623 return -EBADFD;
9624
9625 if (ctx->restrictions.registered)
9626 ctx->restricted = 1;
9627
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009628 io_sq_offload_start(ctx);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009629 return 0;
9630}
9631
Jens Axboe071698e2020-01-28 10:04:42 -07009632static bool io_register_op_must_quiesce(int op)
9633{
9634 switch (op) {
9635 case IORING_UNREGISTER_FILES:
9636 case IORING_REGISTER_FILES_UPDATE:
9637 case IORING_REGISTER_PROBE:
9638 case IORING_REGISTER_PERSONALITY:
9639 case IORING_UNREGISTER_PERSONALITY:
9640 return false;
9641 default:
9642 return true;
9643 }
9644}
9645
Jens Axboeedafcce2019-01-09 09:16:05 -07009646static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
9647 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -06009648 __releases(ctx->uring_lock)
9649 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -07009650{
9651 int ret;
9652
Jens Axboe35fa71a2019-04-22 10:23:23 -06009653 /*
9654 * We're inside the ring mutex, if the ref is already dying, then
9655 * someone else killed the ctx or is already going through
9656 * io_uring_register().
9657 */
9658 if (percpu_ref_is_dying(&ctx->refs))
9659 return -ENXIO;
9660
Jens Axboe071698e2020-01-28 10:04:42 -07009661 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009662 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -06009663
Jens Axboe05f3fb32019-12-09 11:22:50 -07009664 /*
9665 * Drop uring mutex before waiting for references to exit. If
9666 * another thread is currently inside io_uring_enter() it might
9667 * need to grab the uring_lock to make progress. If we hold it
9668 * here across the drain wait, then we can deadlock. It's safe
9669 * to drop the mutex here, since no new references will come in
9670 * after we've killed the percpu ref.
9671 */
9672 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009673 do {
9674 ret = wait_for_completion_interruptible(&ctx->ref_comp);
9675 if (!ret)
9676 break;
Jens Axboeed6930c2020-10-08 19:09:46 -06009677 ret = io_run_task_work_sig();
9678 if (ret < 0)
9679 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009680 } while (1);
9681
Jens Axboe05f3fb32019-12-09 11:22:50 -07009682 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009683
Jens Axboec1503682020-01-08 08:26:07 -07009684 if (ret) {
9685 percpu_ref_resurrect(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009686 goto out_quiesce;
9687 }
9688 }
9689
9690 if (ctx->restricted) {
9691 if (opcode >= IORING_REGISTER_LAST) {
9692 ret = -EINVAL;
9693 goto out;
9694 }
9695
9696 if (!test_bit(opcode, ctx->restrictions.register_op)) {
9697 ret = -EACCES;
Jens Axboec1503682020-01-08 08:26:07 -07009698 goto out;
9699 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07009700 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009701
9702 switch (opcode) {
9703 case IORING_REGISTER_BUFFERS:
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009704 ret = io_sqe_buffers_register(ctx, arg, nr_args);
Jens Axboeedafcce2019-01-09 09:16:05 -07009705 break;
9706 case IORING_UNREGISTER_BUFFERS:
9707 ret = -EINVAL;
9708 if (arg || nr_args)
9709 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009710 ret = io_sqe_buffers_unregister(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -07009711 break;
Jens Axboe6b063142019-01-10 22:13:58 -07009712 case IORING_REGISTER_FILES:
9713 ret = io_sqe_files_register(ctx, arg, nr_args);
9714 break;
9715 case IORING_UNREGISTER_FILES:
9716 ret = -EINVAL;
9717 if (arg || nr_args)
9718 break;
9719 ret = io_sqe_files_unregister(ctx);
9720 break;
Jens Axboec3a31e62019-10-03 13:59:56 -06009721 case IORING_REGISTER_FILES_UPDATE:
9722 ret = io_sqe_files_update(ctx, arg, nr_args);
9723 break;
Jens Axboe9b402842019-04-11 11:45:41 -06009724 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -07009725 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -06009726 ret = -EINVAL;
9727 if (nr_args != 1)
9728 break;
9729 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -07009730 if (ret)
9731 break;
9732 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
9733 ctx->eventfd_async = 1;
9734 else
9735 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -06009736 break;
9737 case IORING_UNREGISTER_EVENTFD:
9738 ret = -EINVAL;
9739 if (arg || nr_args)
9740 break;
9741 ret = io_eventfd_unregister(ctx);
9742 break;
Jens Axboe66f4af92020-01-16 15:36:52 -07009743 case IORING_REGISTER_PROBE:
9744 ret = -EINVAL;
9745 if (!arg || nr_args > 256)
9746 break;
9747 ret = io_probe(ctx, arg, nr_args);
9748 break;
Jens Axboe071698e2020-01-28 10:04:42 -07009749 case IORING_REGISTER_PERSONALITY:
9750 ret = -EINVAL;
9751 if (arg || nr_args)
9752 break;
9753 ret = io_register_personality(ctx);
9754 break;
9755 case IORING_UNREGISTER_PERSONALITY:
9756 ret = -EINVAL;
9757 if (arg)
9758 break;
9759 ret = io_unregister_personality(ctx, nr_args);
9760 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009761 case IORING_REGISTER_ENABLE_RINGS:
9762 ret = -EINVAL;
9763 if (arg || nr_args)
9764 break;
9765 ret = io_register_enable_rings(ctx);
9766 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009767 case IORING_REGISTER_RESTRICTIONS:
9768 ret = io_register_restrictions(ctx, arg, nr_args);
9769 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07009770 default:
9771 ret = -EINVAL;
9772 break;
9773 }
9774
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009775out:
Jens Axboe071698e2020-01-28 10:04:42 -07009776 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009777 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -07009778 percpu_ref_reinit(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009779out_quiesce:
Jens Axboe0f158b42020-05-14 17:18:39 -06009780 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009781 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009782 return ret;
9783}
9784
9785SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
9786 void __user *, arg, unsigned int, nr_args)
9787{
9788 struct io_ring_ctx *ctx;
9789 long ret = -EBADF;
9790 struct fd f;
9791
9792 f = fdget(fd);
9793 if (!f.file)
9794 return -EBADF;
9795
9796 ret = -EOPNOTSUPP;
9797 if (f.file->f_op != &io_uring_fops)
9798 goto out_fput;
9799
9800 ctx = f.file->private_data;
9801
Pavel Begunkovb6c23dd2021-02-20 15:17:18 +00009802 io_run_task_work();
9803
Jens Axboeedafcce2019-01-09 09:16:05 -07009804 mutex_lock(&ctx->uring_lock);
9805 ret = __io_uring_register(ctx, opcode, arg, nr_args);
9806 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009807 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
9808 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07009809out_fput:
9810 fdput(f);
9811 return ret;
9812}
9813
Jens Axboe2b188cc2019-01-07 10:46:33 -07009814static int __init io_uring_init(void)
9815{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009816#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
9817 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
9818 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
9819} while (0)
9820
9821#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
9822 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
9823 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
9824 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
9825 BUILD_BUG_SQE_ELEM(1, __u8, flags);
9826 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
9827 BUILD_BUG_SQE_ELEM(4, __s32, fd);
9828 BUILD_BUG_SQE_ELEM(8, __u64, off);
9829 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
9830 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009831 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009832 BUILD_BUG_SQE_ELEM(24, __u32, len);
9833 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
9834 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
9835 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
9836 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +08009837 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
9838 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009839 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
9840 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
9841 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
9842 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
9843 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
9844 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
9845 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
9846 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009847 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009848 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
9849 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
9850 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009851 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009852
Jens Axboed3656342019-12-18 09:50:26 -07009853 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -07009854 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Jens Axboe91f245d2021-02-09 13:48:50 -07009855 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
9856 SLAB_ACCOUNT);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009857 return 0;
9858};
9859__initcall(io_uring_init);